Add exported state.

config.json

@@ -0,0 +1,30 @@
+{
+  "_name_or_path": "decapoda-research/llama-65b-hf",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "bos_token_id": 0,
+  "custom_pipelines": {
+    "text-generation": {
+      "impl": "h2oai_pipeline.H2OTextGenerationPipeline",
+      "pt": "AutoModelForCausalLM"
+    }
+  },
+  "eos_token_id": 1,
+  "hidden_act": "silu",
+  "hidden_size": 8192,
+  "initializer_range": 0.02,
+  "intermediate_size": 22016,
+  "max_position_embeddings": 2048,
+  "max_sequence_length": 2048,
+  "model_type": "llama",
+  "num_attention_heads": 64,
+  "num_hidden_layers": 80,
+  "pad_token_id": -1,
+  "rms_norm_eps": 1e-05,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.30.1",
+  "use_cache": true,
+  "vocab_size": 32000
+}

generation_config.json

@@ -0,0 +1,7 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 0,
+  "eos_token_id": 1,
+  "pad_token_id": 0,
+  "transformers_version": "4.30.1"
+}

h2oai_pipeline.py

@@ -0,0 +1,929 @@
+import os
+
+from transformers import TextGenerationPipeline
+from transformers.pipelines.text_generation import ReturnType
+
+
+
+
+
+class H2OTextGenerationPipeline(TextGenerationPipeline):
+    def __init__(self, *args, debug=False, chat=False, stream_output=False,
+                 sanitize_bot_response=False,
+                 use_prompter=True, prompter=None,
+                 prompt_type=None, prompt_dict=None,
+                 max_input_tokens=2048 - 256, **kwargs):
+        """
+        HF-like pipeline, but handle instruction prompting and stopping (for some models)
+        :param args:
+        :param debug:
+        :param chat:
+        :param stream_output:
+        :param sanitize_bot_response:
+        :param use_prompter: Whether to use prompter.  If pass prompt_type, will make prompter
+        :param prompter: prompter, can pass if have already
+        :param prompt_type: prompt_type, e.g. human_bot.  See prompt_type to model mapping in 
+                            If use_prompter, then will make prompter and use it.
+        :param prompt_dict: dict of get_prompt(, return_dict=True) for prompt_type=custom
+        :param max_input_tokens:
+        :param kwargs:
+        """
+        super().__init__(*args, **kwargs)
+        self.prompt_text = None
+        self.use_prompter = use_prompter
+        self.prompt_type = prompt_type
+        self.prompt_dict = prompt_dict
+        self.prompter = prompter
+        if self.use_prompter:
+            if self.prompter is not None:
+                assert self.prompter.prompt_type is not None
+            else:
+                self.prompter = Prompter(self.prompt_type, self.prompt_dict, debug=debug, chat=chat,
+                                         stream_output=stream_output)
+            self.human = self.prompter.humanstr
+            self.bot = self.prompter.botstr
+            self.can_stop = True
+        else:
+            self.prompter = None
+            self.human = None
+            self.bot = None
+            self.can_stop = False
+        self.sanitize_bot_response = sanitize_bot_response
+        self.max_input_tokens = max_input_tokens  # not for generate, so ok that not kwargs
+
+    @staticmethod
+    def limit_prompt(prompt_text, tokenizer, max_prompt_length=None):
+        verbose = bool(int(os.getenv('VERBOSE_PIPELINE', '0')))
+
+        if hasattr(tokenizer, 'model_max_length'):
+            # model_max_length only defined for generate.py, not raw use of h2oai_pipeline.py
+            model_max_length = tokenizer.model_max_length
+            if max_prompt_length is not None:
+                model_max_length = min(model_max_length, max_prompt_length)
+            # cut at some upper likely limit to avoid excessive tokenization etc
+            # upper bound of 10 chars/token, e.g. special chars sometimes are long
+            if len(prompt_text) > model_max_length * 10:
+                len0 = len(prompt_text)
+                prompt_text = prompt_text[-model_max_length * 10:]
+                if verbose:
+                    print("Cut of input: %s -> %s" % (len0, len(prompt_text)), flush=True)
+        else:
+            # unknown
+            model_max_length = None
+
+        if model_max_length is not None:
+            num_prompt_tokens = None
+            # can't wait for "hole" if not plain prompt_type, since would lose prefix like <human>:
+            # For https://github.com/h2oai/h2ogpt/issues/192
+            for trial in range(0, 3):
+                prompt_tokens = tokenizer(prompt_text)['input_ids']
+                num_prompt_tokens = len(prompt_tokens)
+                if num_prompt_tokens > model_max_length:
+                    # conservative by using int()
+                    chars_per_token = int(len(prompt_text) / num_prompt_tokens)
+                    # keep tail, where question is if using langchain
+                    prompt_text = prompt_text[-model_max_length * chars_per_token:]
+                    if verbose:
+                        print("reducing %s tokens, assuming average of %s chars/token for %s characters" % (
+                            num_prompt_tokens, chars_per_token, len(prompt_text)), flush=True)
+                else:
+                    if verbose:
+                        print("using %s tokens with %s chars" % (num_prompt_tokens, len(prompt_text)), flush=True)
+                    break
+
+            # Why Below False: don't limit max_new_tokens more, just rely upon stopping to reach limit of model
+            if False:
+                # if input prompt is some number of tokens, despite user request, can't have max_new_tokens more
+                #
+                assert num_prompt_tokens is not None
+                if self.prompt_type not in [PromptType.plain.name, PromptType.plain.value]:
+                    # then give room for prompt
+                    fudge = 20
+                else:
+                    fudge = 0
+                max_new_tokens = max(0, min(generate_kwargs['max_new_tokens'],
+                                            model_max_length - (num_prompt_tokens + fudge)))
+                if max_new_tokens < generate_kwargs['max_new_tokens']:
+                    if verbose:
+                        print("Reduced max_new_tokens from %s -> %s" % (
+                        generate_kwargs['max_new_tokens'], max_new_tokens))
+                    generate_kwargs['max_new_tokens'] = max_new_tokens
+        return prompt_text
+
+    def preprocess(self, prompt_text, prefix="", handle_long_generation=None, **generate_kwargs):
+        prompt_text = H2OTextGenerationPipeline.limit_prompt(prompt_text, self.tokenizer)
+
+        data_point = dict(context='', instruction=prompt_text, input='')
+        if self.prompter is not None:
+            prompt_text = self.prompter.generate_prompt(data_point)
+        self.prompt_text = prompt_text
+        if handle_long_generation is None:
+            # forces truncation of inputs to avoid critical failure
+            handle_long_generation = None  # disable with new approaches
+        return super().preprocess(prompt_text, prefix=prefix, handle_long_generation=handle_long_generation,
+                                  **generate_kwargs)
+
+    def postprocess(self, model_outputs, return_type=ReturnType.FULL_TEXT, clean_up_tokenization_spaces=True):
+        records = super().postprocess(model_outputs, return_type=return_type,
+                                      clean_up_tokenization_spaces=clean_up_tokenization_spaces)
+        for rec in records:
+            if self.use_prompter:
+                outputs = rec['generated_text']
+                outputs = self.prompter.get_response(outputs, prompt=self.prompt_text,
+                                                     sanitize_bot_response=self.sanitize_bot_response)
+            elif self.bot and self.human:
+                outputs = rec['generated_text'].split(self.bot)[1].strip().split(self.human)[0].strip()
+            else:
+                outputs = rec['generated_text']
+            rec['generated_text'] = outputs
+        return records
+
+    def _forward(self, model_inputs, **generate_kwargs):
+        if self.can_stop:
+            stopping_criteria = get_stopping(self.prompt_type, self.prompt_dict,
+                                             self.tokenizer, self.device,
+                                             human=self.human, bot=self.bot,
+                                             model_max_length=self.tokenizer.model_max_length)
+            generate_kwargs['stopping_criteria'] = stopping_criteria
+        # return super()._forward(model_inputs, **generate_kwargs)
+        return self.__forward(model_inputs, **generate_kwargs)
+
+    # FIXME: Copy-paste of original _forward, but removed copy.deepcopy()
+    # FIXME: https://github.com/h2oai/h2ogpt/issues/172
+    def __forward(self, model_inputs, **generate_kwargs):
+        input_ids = model_inputs["input_ids"]
+        attention_mask = model_inputs.get("attention_mask", None)
+        # Allow empty prompts
+        if input_ids.shape[1] == 0:
+            input_ids = None
+            attention_mask = None
+            in_b = 1
+        else:
+            in_b = input_ids.shape[0]
+        prompt_text = model_inputs.pop("prompt_text")
+
+        ## If there is a prefix, we may need to adjust the generation length. Do so without permanently modifying
+        ## generate_kwargs, as some of the parameterization may come from the initialization of the pipeline.
+        # generate_kwargs = copy.deepcopy(generate_kwargs)
+        prefix_length = generate_kwargs.pop("prefix_length", 0)
+        if prefix_length > 0:
+            has_max_new_tokens = "max_new_tokens" in generate_kwargs or (
+                    "generation_config" in generate_kwargs
+                    and generate_kwargs["generation_config"].max_new_tokens is not None
+            )
+            if not has_max_new_tokens:
+                generate_kwargs["max_length"] = generate_kwargs.get("max_length") or self.model.config.max_length
+                generate_kwargs["max_length"] += prefix_length
+            has_min_new_tokens = "min_new_tokens" in generate_kwargs or (
+                    "generation_config" in generate_kwargs
+                    and generate_kwargs["generation_config"].min_new_tokens is not None
+            )
+            if not has_min_new_tokens and "min_length" in generate_kwargs:
+                generate_kwargs["min_length"] += prefix_length
+
+        # BS x SL
+        generated_sequence = self.model.generate(input_ids=input_ids, attention_mask=attention_mask, **generate_kwargs)
+        out_b = generated_sequence.shape[0]
+        if self.framework == "pt":
+            generated_sequence = generated_sequence.reshape(in_b, out_b // in_b, *generated_sequence.shape[1:])
+        elif self.framework == "tf":
+            from transformers import is_tf_available
+            if is_tf_available():
+                import tensorflow as tf
+                generated_sequence = tf.reshape(generated_sequence,
+                                                (in_b, out_b // in_b, *generated_sequence.shape[1:]))
+            else:
+                raise ValueError("TF not avaialble.")
+        return {"generated_sequence": generated_sequence, "input_ids": input_ids, "prompt_text": prompt_text}
+import torch
+from transformers import StoppingCriteria, StoppingCriteriaList
+
+
+
+class StoppingCriteriaSub(StoppingCriteria):
+
+    def __init__(self, stops=[], encounters=[], device="cuda", model_max_length=None):
+        super().__init__()
+        assert len(stops) % len(encounters) == 0, "Number of stops and encounters must match"
+        self.encounters = encounters
+        self.stops = [stop.to(device) for stop in stops]
+        self.num_stops = [0] * len(stops)
+        self.model_max_length = model_max_length
+
+    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
+        for stopi, stop in enumerate(self.stops):
+            if torch.all((stop == input_ids[0][-len(stop):])).item():
+                self.num_stops[stopi] += 1
+                if self.num_stops[stopi] >= self.encounters[stopi % len(self.encounters)]:
+                    # print("Stopped", flush=True)
+                    return True
+        if self.model_max_length is not None and input_ids[0].shape[0] >= self.model_max_length:
+            # critical limit
+            return True
+        # print("Tokens: %s" % input_ids[0].cpu().numpy(), flush=True)
+        # print("Stop Tokens: %s" % [x.cpu().numpy() for x in self.stops], flush=True)
+        return False
+
+
+def get_stopping(prompt_type, prompt_dict, tokenizer, device, human='<human>:', bot="<bot>:", model_max_length=None):
+    # FIXME: prompt_dict unused currently
+    if prompt_type in [PromptType.human_bot.name, PromptType.instruct_vicuna.name, PromptType.instruct_with_end.name]:
+        if prompt_type == PromptType.human_bot.name:
+            # encounters = [prompt.count(human) + 1, prompt.count(bot) + 1]
+            # stopping only starts once output is beyond prompt
+            # 1 human is enough to trigger, but need 2 bots, because very first view back will be bot we added
+            stop_words = [human, bot, '\n' + human, '\n' + bot]
+            encounters = [1, 2]
+        elif prompt_type == PromptType.instruct_vicuna.name:
+            # even below is not enough, generic strings and many ways to encode
+            stop_words = [
+                '### Human:',
+                """
+### Human:""",
+                """
+### Human:
+""",
+                '### Assistant:',
+                """
+### Assistant:""",
+                """
+### Assistant:
+""",
+            ]
+            encounters = [1, 2]
+        else:
+            # some instruct prompts have this as end, doesn't hurt to stop on it since not common otherwise
+            stop_words = ['### End']
+            encounters = [1]
+        stop_words_ids = [
+            tokenizer(stop_word, return_tensors='pt')['input_ids'].squeeze() for stop_word in stop_words]
+        # handle single token case
+        stop_words_ids = [x if len(x.shape) > 0 else torch.tensor([x]) for x in stop_words_ids]
+        stop_words_ids = [x for x in stop_words_ids if x.shape[0] > 0]
+        # avoid padding in front of tokens
+        if tokenizer._pad_token:  # use hidden variable to avoid annoying properly logger bug
+            stop_words_ids = [x[1:] if x[0] == tokenizer.pad_token_id and len(x) > 1 else x for x in stop_words_ids]
+        # handle fake \n added
+        stop_words_ids = [x[1:] if y[0] == '\n' else x for x, y in zip(stop_words_ids, stop_words)]
+        # build stopper
+        stopping_criteria = StoppingCriteriaList(
+            [StoppingCriteriaSub(stops=stop_words_ids, encounters=encounters, device=device,
+                                 model_max_length=model_max_length)])
+    else:
+        stopping_criteria = StoppingCriteriaList()
+    return stopping_criteria
+from enum import Enum
+
+
+class PromptType(Enum):
+    custom = -1
+    plain = 0
+    instruct = 1
+    quality = 2
+    human_bot = 3
+    dai_faq = 4
+    summarize = 5
+    simple_instruct = 6
+    instruct_vicuna = 7
+    instruct_with_end = 8
+    human_bot_orig = 9
+    prompt_answer = 10
+    open_assistant = 11
+    wizard_lm = 12
+    wizard_mega = 13
+    instruct_vicuna2 = 14
+    instruct_vicuna3 = 15
+    wizard2 = 16
+    wizard3 = 17
+    instruct_simple = 18
+
+
+class DocumentChoices(Enum):
+    All_Relevant = 0
+    All_Relevant_Only_Sources = 1
+    Only_All_Sources = 2
+    Just_LLM = 3
+
+
+class LangChainMode(Enum):
+    """LangChain mode"""
+
+    DISABLED = "Disabled"
+    CHAT_LLM = "ChatLLM"
+    LLM = "LLM"
+    ALL = "All"
+    WIKI = "wiki"
+    WIKI_FULL = "wiki_full"
+    USER_DATA = "UserData"
+    MY_DATA = "MyData"
+    GITHUB_H2OGPT = "github h2oGPT"
+    H2O_DAI_DOCS = "DriverlessAI docs"
+import ast
+import time
+from enums import PromptType  # also supports imports from this file from other files
+
+non_hf_types = ['gpt4all_llama', 'llama', 'gptj']
+
+prompt_type_to_model_name = {
+    'plain': [
+        'EleutherAI/gpt-j-6B',
+        'EleutherAI/pythia-6.9b',
+        'EleutherAI/pythia-12b',
+        'EleutherAI/pythia-12b-deduped',
+        'EleutherAI/gpt-neox-20b',
+        'openlm-research/open_llama_7b_700bt_preview',
+        'decapoda-research/llama-7b-hf',
+        'decapoda-research/llama-13b-hf',
+        'decapoda-research/llama-30b-hf',
+        'decapoda-research/llama-65b-hf',
+        'facebook/mbart-large-50-many-to-many-mmt',
+        'philschmid/bart-large-cnn-samsum',
+        'philschmid/flan-t5-base-samsum',
+        'gpt2',
+        'distilgpt2',
+        'mosaicml/mpt-7b-storywriter',
+        'mosaicml/mpt-7b-instruct',  # internal code handles instruct
+        'mosaicml/mpt-7b-chat',  # NC, internal code handles instruct
+        'gptj',  # internally handles prompting
+        'llama',  # plain, or need to choose prompt_type for given TheBloke model
+        'gpt4all_llama',  # internally handles prompting
+    ],
+    'prompt_answer': [
+        'h2oai/h2ogpt-gm-oasst1-en-1024-20b',
+        'h2oai/h2ogpt-gm-oasst1-en-1024-12b',
+        'h2oai/h2ogpt-gm-oasst1-multilang-1024-20b',
+        'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b-preview-300bt',
+        'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b-preview-300bt-v2',
+        'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b-preview-700bt',
+        'h2oai/h2ogpt-gm-oasst1-en-2048-open-llama-7b',
+        'h2oai/h2ogpt-gm-oasst1-multilang-2048-falcon-7b',
+        'h2oai/h2ogpt-gm-oasst1-multilang-2048-falcon-7b-v2',
+        'h2oai/h2ogpt-gm-oasst1-en-2048-falcon-7b',
+        'h2oai/h2ogpt-gm-oasst1-en-2048-falcon-7b-v2',
+        'h2oai/h2ogpt-gm-oasst1-en-2048-falcon-40b-v1',
+    ],
+    'instruct': [],
+    'instruct_with_end': ['databricks/dolly-v2-12b'],
+    'quality': [],
+    'human_bot': [
+        'h2oai/h2ogpt-oasst1-512-12b',
+        'h2oai/h2ogpt-oasst1-512-20b',
+        'h2oai/h2ogpt-oig-oasst1-256-6_9b',
+        'h2oai/h2ogpt-oig-oasst1-512-6_9b',
+        'h2oai/h2ogpt-oig-oasst1-256-6.9b',  # legacy
+        'h2oai/h2ogpt-oig-oasst1-512-6.9b',  # legacy
+        'h2oai/h2ogpt-research-oasst1-512-30b',
+        'h2oai/h2ogpt-oasst1-falcon-40b',
+        'h2oai/h2ogpt-oig-oasst1-falcon-40b',
+    ],
+    'dai_faq': [],
+    'summarize': [],
+    'simple_instruct': ['t5-small', 't5-large', 'google/flan-t5', 'google/flan-t5-xxl', 'google/flan-ul2'],
+    'instruct_vicuna': ['AlekseyKorshuk/vicuna-7b', 'TheBloke/stable-vicuna-13B-HF', 'junelee/wizard-vicuna-13b'],
+    'human_bot_orig': ['togethercomputer/GPT-NeoXT-Chat-Base-20B'],
+    "open_assistant": ['OpenAssistant/oasst-sft-7-llama-30b-xor', 'oasst-sft-7-llama-30b'],
+    "wizard_lm": ['ehartford/WizardLM-7B-Uncensored', 'ehartford/WizardLM-13B-Uncensored'],
+    "wizard_mega": ['openaccess-ai-collective/wizard-mega-13b'],
+    "instruct_simple": ['JosephusCheung/Guanaco'],
+}
+
+inv_prompt_type_to_model_name = {v.strip(): k for k, l in prompt_type_to_model_name.items() for v in l}
+inv_prompt_type_to_model_lower = {v.strip().lower(): k for k, l in prompt_type_to_model_name.items() for v in l}
+
+prompt_types_strings = []
+for p in PromptType:
+    prompt_types_strings.extend([p.name])
+
+prompt_types = []
+for p in PromptType:
+    prompt_types.extend([p.name, p.value, str(p.value)])
+
+
+def get_prompt(prompt_type, prompt_dict, chat, context, reduced, return_dict=False):
+    prompt_dict_error = ''
+    if prompt_type == PromptType.custom.name and not isinstance(prompt_dict, dict):
+        try:
+            prompt_dict = ast.literal_eval(prompt_dict)
+        except BaseException as e:
+            prompt_dict_error = str(e)
+        if prompt_dict_error:
+            return dict(), prompt_dict_error
+
+    if prompt_type in [PromptType.custom.value, str(PromptType.custom.value),
+                       PromptType.custom.name]:
+        promptA = prompt_dict.get('promptA', '')
+        promptB = prompt_dict('promptB', '')
+        PreInstruct = prompt_dict.get('PreInstruct', '')
+        PreInput = prompt_dict.get('PreInput', '')
+        PreResponse = prompt_dict.get('PreResponse', '')
+        terminate_response = prompt_dict.get('terminate_response', None)
+        chat_sep = prompt_dict.get('chat_sep', '\n')
+        humanstr = prompt_dict.get('humanstr', '')
+        botstr = prompt_dict.get('botstr', '')
+    elif prompt_type in [PromptType.plain.value, str(PromptType.plain.value),
+                         PromptType.plain.name]:
+        promptA = promptB = PreInstruct = PreInput = PreResponse = ''
+        terminate_response = []
+        chat_sep = ''
+        humanstr = ''
+        botstr = ''
+    elif prompt_type == 'simple_instruct':
+        promptA = promptB = PreInstruct = PreInput = PreResponse = None
+        terminate_response = []
+        chat_sep = '\n'
+        humanstr = ''
+        botstr = ''
+    elif prompt_type in [PromptType.instruct.value, str(PromptType.instruct.value),
+                         PromptType.instruct.name] + [PromptType.instruct_with_end.value,
+                                                      str(PromptType.instruct_with_end.value),
+                                                      PromptType.instruct_with_end.name]:
+        promptA = 'Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\n' if not (
+                chat and reduced) else ''
+        promptB = 'Below is an instruction that describes a task. Write a response that appropriately completes the request.\n' if not (
+                chat and reduced) else ''
+
+        PreInstruct = """
+### Instruction:
+"""
+
+        PreInput = """
+### Input:
+"""
+
+        PreResponse = """
+### Response:
+"""
+        if prompt_type in [PromptType.instruct_with_end.value, str(PromptType.instruct_with_end.value),
+                           PromptType.instruct_with_end.name]:
+            terminate_response = ['### End']
+        else:
+            terminate_response = None
+        chat_sep = '\n'
+        humanstr = PreInstruct
+        botstr = PreResponse
+    elif prompt_type in [PromptType.quality.value, str(PromptType.quality.value),
+                         PromptType.quality.name]:
+        promptA = 'Write a detailed high-quality, accurate, fair, Response with about 100 words by following the Instruction as applied on the Input.\n' if not (
+                chat and reduced) else ''
+        promptB = 'Write a detailed high-quality, accurate, fair, Response with about 100 words by following the Instruction.\n' if not (
+                chat and reduced) else ''
+
+        PreInstruct = """
+### Instruction:
+"""
+
+        PreInput = """
+### Input:
+"""
+
+        PreResponse = """
+### Response:
+"""
+        terminate_response = None
+        chat_sep = '\n'
+        humanstr = PreInstruct  # first thing human says
+        botstr = PreResponse  # first thing bot says
+    elif prompt_type in [PromptType.human_bot.value, str(PromptType.human_bot.value),
+                         PromptType.human_bot.name] + [PromptType.human_bot_orig.value,
+                                                       str(PromptType.human_bot_orig.value),
+                                                       PromptType.human_bot_orig.name]:
+        human = '<human>:'
+        bot = "<bot>:"
+        if reduced or context or prompt_type in [PromptType.human_bot.value, str(PromptType.human_bot.value),
+                                                 PromptType.human_bot.name]:
+            preprompt = ''
+        else:
+            cur_date = time.strftime('%Y-%m-%d')
+            cur_time = time.strftime('%H:%M:%S %p %Z')
+
+            PRE_PROMPT = """\
+Current Date: {}
+Current Time: {}
+
+"""
+            preprompt = PRE_PROMPT.format(cur_date, cur_time)
+        start = human
+        promptB = promptA = '%s%s ' % (preprompt, start)
+
+        PreInstruct = ""
+
+        PreInput = None
+
+        if reduced:
+            # when making context, want it to appear as-if LLM generated, which starts with space after :
+            PreResponse = bot + ' '
+        else:
+            # normally LLM adds space after this, because was how trained.
+            # if add space here, non-unique tokenization will often make LLM produce wrong output
+            PreResponse = bot
+
+        terminate_response = [start, PreResponse]
+        chat_sep = '\n'
+        humanstr = human  # tag before human talks
+        botstr = bot  # tag before bot talks
+    elif prompt_type in [PromptType.dai_faq.value, str(PromptType.dai_faq.value),
+                         PromptType.dai_faq.name]:
+        promptA = ''
+        promptB = 'Answer the following Driverless AI question.\n'
+
+        PreInstruct = """
+### Driverless AI frequently asked question:
+"""
+
+        PreInput = None
+
+        PreResponse = """
+### Driverless AI documentation answer:
+"""
+        terminate_response = ['\n\n']
+        chat_sep = terminate_response
+        humanstr = PreInstruct
+        botstr = PreResponse
+    elif prompt_type in [PromptType.summarize.value, str(PromptType.summarize.value),
+                         PromptType.summarize.name]:
+        promptA = promptB = PreInput = ''
+        PreInstruct = '## Main Text\n\n'
+        PreResponse = '\n\n## Summary\n\n'
+        terminate_response = None
+        chat_sep = '\n'
+        humanstr = PreInstruct
+        botstr = PreResponse
+    elif prompt_type in [PromptType.instruct_vicuna.value, str(PromptType.instruct_vicuna.value),
+                         PromptType.instruct_vicuna.name]:
+        promptA = promptB = "A chat between a curious human and an artificial intelligence assistant. " \
+                            "The assistant gives helpful, detailed, and polite answers to the human's questions." if not (
+                chat and reduced) else ''
+
+        PreInstruct = """
+### Human:
+"""
+
+        PreInput = None
+
+        PreResponse = """
+### Assistant:
+"""
+        terminate_response = [
+            '### Human:']  # but only allow terminate after prompt is found correctly, else can't terminate
+        chat_sep = '\n'
+        humanstr = PreInstruct
+        botstr = PreResponse
+    elif prompt_type in [PromptType.prompt_answer.value, str(PromptType.prompt_answer.value),
+                         PromptType.prompt_answer.name]:
+        preprompt = ''
+        prompt_tokens = "<|prompt|>"
+        answer_tokens = "<|answer|>"
+        start = prompt_tokens
+        promptB = promptA = '%s%s' % (preprompt, start)
+        PreInstruct = ""
+        PreInput = None
+        PreResponse = answer_tokens
+        eos = '<|endoftext|>'  # neox eos
+        terminate_response = [start, PreResponse, eos]
+        chat_sep = eos
+        humanstr = prompt_tokens
+        botstr = answer_tokens
+    elif prompt_type in [PromptType.open_assistant.value, str(PromptType.open_assistant.value),
+                         PromptType.open_assistant.name]:
+        # From added_tokens.json
+        preprompt = ''
+        prompt_tokens = "<|prompter|>"
+        answer_tokens = "<|assistant|>"
+        start = prompt_tokens
+        promptB = promptA = '%s%s' % (preprompt, start)
+        PreInstruct = ""
+        PreInput = None
+        PreResponse = answer_tokens
+        pend = "<|prefix_end|>"
+        eos = "</s>"
+        terminate_response = [start, PreResponse, pend, eos]
+        chat_sep = eos
+        humanstr = prompt_tokens
+        botstr = answer_tokens
+    elif prompt_type in [PromptType.wizard_lm.value, str(PromptType.wizard_lm.value),
+                         PromptType.wizard_lm.name]:
+        # https://github.com/ehartford/WizardLM/blob/main/src/train_freeform.py
+        preprompt = ''
+        start = ''
+        promptB = promptA = '%s%s' % (preprompt, start)
+        PreInstruct = ""
+        PreInput = None
+        PreResponse = "\n\n### Response\n"
+        eos = "</s>"
+        terminate_response = [PreResponse, eos]
+        chat_sep = eos
+        humanstr = promptA
+        botstr = PreResponse
+    elif prompt_type in [PromptType.wizard_mega.value, str(PromptType.wizard_mega.value),
+                         PromptType.wizard_mega.name]:
+        preprompt = ''
+        start = ''
+        promptB = promptA = '%s%s' % (preprompt, start)
+        PreInstruct = """
+### Instruction:
+"""
+        PreInput = None
+        PreResponse = """
+### Assistant:
+"""
+        terminate_response = [PreResponse]
+        chat_sep = '\n'
+        humanstr = PreInstruct
+        botstr = PreResponse
+    elif prompt_type in [PromptType.instruct_vicuna2.value, str(PromptType.instruct_vicuna2.value),
+                         PromptType.instruct_vicuna2.name]:
+        promptA = promptB = "" if not (
+                chat and reduced) else ''
+
+        PreInstruct = """
+HUMAN:
+"""
+
+        PreInput = None
+
+        PreResponse = """
+ASSISTANT:
+"""
+        terminate_response = [
+            'HUMAN:']  # but only allow terminate after prompt is found correctly, else can't terminate
+        chat_sep = '\n'
+        humanstr = PreInstruct
+        botstr = PreResponse
+    elif prompt_type in [PromptType.instruct_vicuna3.value, str(PromptType.instruct_vicuna3.value),
+                         PromptType.instruct_vicuna3.name]:
+        promptA = promptB = "" if not (
+                chat and reduced) else ''
+
+        PreInstruct = """
+### User:
+"""
+
+        PreInput = None
+
+        PreResponse = """
+### Assistant:
+"""
+        terminate_response = [
+            '### User:']  # but only allow terminate after prompt is found correctly, else can't terminate
+        chat_sep = '\n'
+        humanstr = PreInstruct
+        botstr = PreResponse
+    elif prompt_type in [PromptType.wizard2.value, str(PromptType.wizard2.value),
+                         PromptType.wizard2.name]:
+        # https://huggingface.co/TheBloke/WizardLM-7B-uncensored-GGML
+        preprompt = """Below is an instruction that describes a task. Write a response that appropriately completes the request."""
+        start = ''
+        promptB = promptA = '%s%s' % (preprompt, start)
+        PreInstruct = """
+### Instruction:
+"""
+        PreInput = None
+        PreResponse = """
+### Response:
+"""
+        terminate_response = [PreResponse]
+        chat_sep = '\n'
+        humanstr = PreInstruct
+        botstr = PreResponse
+    elif prompt_type in [PromptType.wizard3.value, str(PromptType.wizard3.value),
+                         PromptType.wizard3.name]:
+        # https://huggingface.co/TheBloke/wizardLM-13B-1.0-GGML
+        preprompt = """A chat between a curious user and an artificial intelligence assistant. The assistant gives helpful, detailed, and polite answers to the user's questions."""
+        start = ''
+        promptB = promptA = '%s%s' % (preprompt, start)
+        PreInstruct = """USER: """
+        PreInput = None
+        PreResponse = """ASSISTANT: """
+        terminate_response = [PreResponse]
+        chat_sep = '\n'
+        humanstr = PreInstruct
+        botstr = PreResponse
+
+    elif prompt_type in [PromptType.instruct_simple.value, str(PromptType.instruct_simple.value),
+                         PromptType.instruct_simple.name]:
+        promptA = '' if not (chat and reduced) else ''
+        promptB = '' if not (chat and reduced) else ''
+
+        PreInstruct = """
+### Instruction:
+"""
+
+        PreInput = """
+### Input:
+"""
+
+        PreResponse = """
+### Response:
+"""
+        terminate_response = None
+        chat_sep = '\n'
+        humanstr = PreInstruct
+        botstr = PreResponse
+    else:
+        raise RuntimeError("No such prompt_type=%s" % prompt_type)
+
+    if return_dict:
+        return dict(promptA=promptA, promptB=promptB, PreInstruct=PreInstruct, PreInput=PreInput,
+                    PreResponse=PreResponse, terminate_response=terminate_response, chat_sep=chat_sep,
+                    humanstr=humanstr, botstr=botstr), ''
+    else:
+        return promptA, promptB, PreInstruct, PreInput, PreResponse, terminate_response, chat_sep, humanstr, botstr
+
+
+def generate_prompt(data_point, prompt_type, prompt_dict, chat, reduced):
+    context = data_point.get('context')
+    if context is None:
+        context = ''
+    instruction = data_point.get('instruction')
+    input = data_point.get('input')
+    output = data_point.get('output')
+    prompt_type = data_point.get('prompt_type', prompt_type)
+    prompt_dict = data_point.get('prompt_dict', prompt_dict)
+    assert prompt_type in prompt_types, "Bad prompt type: %s" % prompt_type
+    promptA, promptB, PreInstruct, PreInput, PreResponse, \
+        terminate_response, chat_sep, humanstr, botstr = get_prompt(prompt_type, prompt_dict, chat, context, reduced)
+
+    prompt = context if not reduced else ''
+
+    if input and promptA:
+        prompt += f"""{promptA}"""
+    elif promptB:
+        prompt += f"""{promptB}"""
+
+    if instruction and PreInstruct is not None and input and PreInput is not None:
+        prompt += f"""{PreInstruct}{instruction}{PreInput}{input}"""
+        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
+    elif instruction and input and PreInstruct is None and PreInput is not None:
+        prompt += f"""{PreInput}{instruction}
+{input}"""
+        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
+    elif input and instruction and PreInput is None and PreInstruct is not None:
+        prompt += f"""{PreInstruct}{instruction}
+{input}"""
+        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
+    elif instruction and PreInstruct is not None:
+        prompt += f"""{PreInstruct}{instruction}"""
+        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
+    elif input and PreInput is not None:
+        prompt += f"""{PreInput}{input}"""
+        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
+    elif input and instruction and PreInput is not None:
+        prompt += f"""{PreInput}{instruction}{input}"""
+        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
+    elif input and instruction and PreInstruct is not None:
+        prompt += f"""{PreInstruct}{instruction}{input}"""
+        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
+    elif input and instruction:
+        # i.e. for simple_instruct
+        prompt += f"""{instruction}: {input}"""
+        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
+    elif input:
+        prompt += f"""{input}"""
+        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
+    elif instruction:
+        prompt += f"""{instruction}"""
+        prompt = inject_chatsep(prompt_type, prompt, chat_sep=chat_sep)
+
+    if PreResponse is not None:
+        prompt += f"""{PreResponse}"""
+        pre_response = PreResponse  # Don't use strip
+    else:
+        pre_response = ''
+
+    if output:
+        prompt += f"""{output}"""
+
+    return prompt, pre_response, terminate_response, chat_sep
+
+
+def inject_chatsep(prompt_type, prompt, chat_sep=None):
+    if chat_sep:
+        # only add new line if structured prompt, while 'plain' is just generation of next tokens from input
+        prompt += chat_sep
+    return prompt
+
+
+class Prompter(object):
+    def __init__(self, prompt_type, prompt_dict, debug=False, chat=False, stream_output=False, repeat_penalty=True,
+                 allowed_repeat_line_length=10):
+        self.prompt_type = prompt_type
+        self.prompt_dict = prompt_dict
+        data_point = dict(instruction='', input='', output='')
+        _, self.pre_response, self.terminate_response, self.chat_sep = \
+            generate_prompt(data_point, self.prompt_type, self.prompt_dict, chat, False)
+        self.debug = debug
+        self.chat = chat
+        self.stream_output = stream_output
+        self.repeat_penalty = repeat_penalty
+        self.allowed_repeat_line_length = allowed_repeat_line_length
+        self.prompt = None
+        context = ""  # not for chat context
+        reduced = False  # not for chat context
+        self.promptA, self.promptB, self.PreInstruct, self.PreInput, self.PreResponse, \
+            self.terminate_response, self.chat_sep, self.humanstr, self.botstr = \
+            get_prompt(self.prompt_type, self.prompt_dict, chat, context, reduced)
+
+    def generate_prompt(self, data_point):
+        reduced = False
+        prompt, _, _, _ = generate_prompt(data_point, self.prompt_type, self.prompt_dict, self.chat, reduced)
+        if self.debug:
+            print("prompt: %s" % prompt, flush=True)
+        self.prompt = prompt
+        return prompt
+
+    def get_response(self, outputs, prompt=None, sanitize_bot_response=False):
+        if isinstance(outputs, str):
+            outputs = [outputs]
+        if self.debug:
+            print("output:\n%s" % '\n\n'.join(outputs), flush=True)
+        if prompt is not None:
+            self.prompt = prompt
+
+        def clean_response(response):
+            meaningless_words = ['<pad>', '</s>', '<|endoftext|>']
+            for word in meaningless_words:
+                response = response.replace(word, "")
+            if sanitize_bot_response:
+                from better_profanity import profanity
+                response = profanity.censor(response)
+            response = response.strip("\n")
+            return response
+
+        def clean_repeats(response):
+            lines = response.split('\n')
+            new_lines = []
+            [new_lines.append(line) for line in lines if
+             line not in new_lines or len(line) < self.allowed_repeat_line_length]
+            if self.debug and len(lines) != len(new_lines):
+                print("cleaned repeats: %s %s" % (len(lines), len(new_lines)), flush=True)
+            response = '\n'.join(new_lines)
+            return response
+
+        multi_output = len(outputs) > 1
+
+        for oi, output in enumerate(outputs):
+            if self.prompt_type in [PromptType.plain.value, str(PromptType.plain.value), PromptType.plain.name]:
+                output = clean_response(output)
+            elif prompt is None:
+                # then use most basic parsing like pipeline
+                if self.botstr in output:
+                    if self.humanstr:
+                        output = clean_response(output.split(self.botstr)[1].strip().split(self.humanstr)[0].strip())
+                    else:
+                        # i.e. use after bot but only up to next bot
+                        output = clean_response(output.split(self.botstr)[1].strip().split(self.botstr)[0].strip())
+                else:
+                    # output = clean_response(output.strip())
+                    # assume just not printed yet
+                    output = ""
+            else:
+                # find first instance of prereponse
+                # prompt sometimes has odd characters, that mutate length,
+                # so can't go by length alone
+                if self.pre_response:
+                    outputi = output.find(prompt)
+                    if outputi >= 0:
+                        output = output[outputi + len(prompt):]
+                        allow_terminate = True
+                    else:
+                        # subtraction is risky due to space offsets sometimes, so only do if necessary
+                        output = output[len(prompt) - len(self.pre_response):]
+                        # [1] to avoid repeated pre_response, just take first (after prompt - pre_response for chat)
+                        if self.pre_response in output:
+                            output = output.split(self.pre_response)[1]
+                            allow_terminate = True
+                        else:
+                            if output:
+                                print("Failure of parsing or not enough output yet: %s" % output, flush=True)
+                            allow_terminate = False
+                else:
+                    allow_terminate = True
+                    output = output[len(prompt):]
+                # clean after subtract prompt out, so correct removal of pre_response
+                output = clean_response(output).strip()
+                if self.repeat_penalty:
+                    output = clean_repeats(output).strip()
+                if self.terminate_response and allow_terminate:
+                    finds = []
+                    for term in self.terminate_response:
+                        finds.append(output.find(term))
+                    finds = [x for x in finds if x >= 0]
+                    if len(finds) > 0:
+                        termi = finds[0]
+                        output = output[:termi].strip()
+                    else:
+                        output = output.strip()
+                else:
+                    output = output.strip()
+            if multi_output:
+                # prefix with output counter
+                output = "\n=========== Output %d\n\n" % (1 + oi) + output
+                if oi > 0:
+                    # post fix outputs with seperator
+                    output += '\n'
+            outputs[oi] = output
+        # join all outputs, only one extra new line between outputs
+        output = '\n'.join(outputs)
+        if self.debug:
+            print("outputclean:\n%s" % '\n\n'.join(outputs), flush=True)
+        return output

pytorch_model-00001-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61f6d005b09569252798e4ea185f0c0465e5ab023c5ddf3f94eecb173dce2a86
+size 4659947169

pytorch_model-00002-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20b6abd947d5ed484fdc450038e7926014ac59d9714b65a363006a487f7a0786
+size 4857113187

pytorch_model-00003-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54eaab98e489acb0aa3f57b1458492dc364d2c6c210bd5e12d2352d27eaf8e81
+size 4857113187

pytorch_model-00004-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd58d7e8721393718144677837ac2020e329fbd0d72dc2f5b34c5a95f364a04b
+size 4857113187

pytorch_model-00005-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cca55b5a7c82205dd0e5741c4c5960bafb0b41a360c81f4f2ed793ad490759f3
+size 4857113187

pytorch_model-00006-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eb4f02ec8d41085b127b2639af4b91e8597eae5f1f6781e64d7d121035a0cab4
+size 4857113187

pytorch_model-00007-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b383b8aca0f99a1be5cc3c08a1a955a72ab16d8ae9a23692c18f59a4377138f1
+size 4857113187

pytorch_model-00008-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:031b94889da374a5ca74eb77bcc763158cea24b39358c0a31d9f204837ce487b
+size 4857113187

pytorch_model-00009-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:977a73595c2beacc0f1a75653b0643d145662a9921366aed56b7dfed97dbde11
+size 4857113187

pytorch_model-00010-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:692999f66006f11c518be696bc0a3b545222d88058fa3984ff896f4374b94646
+size 4857113187

pytorch_model-00011-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43870c96f01a50e64a8730538cb12dcfdeb13bbc58830d245cd0a66d28141607
+size 4857113187

pytorch_model-00012-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b104eaa6691efaa55e23e1a9732a4502dfbbdc6a6757fcc49e38984218855b39
+size 4857113187

pytorch_model-00013-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:39c05baec0ebc87567c2308ead7f494b054f3b73dbfac8fac8e6cc8df6ccca68
+size 4857113187

pytorch_model-00014-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8889763ef60c3866e0bfd598b493502e52b3c6b12c35413ae14f1c5b75b922e5
+size 4857113187

pytorch_model-00015-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0a371b03df3047a6b0239681e4155cdd656903ce48d428be984b99bef407b4dc
+size 4857113187

pytorch_model-00016-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ebe52f9719ca301b9c4b28f7ca14bf867138a957243c8cd2d7e89f43146a2893
+size 4857113187

pytorch_model-00017-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3c737acfec6f324c859340cceef04316d891b2267455873765eb02395ddf0e15
+size 4857113187

pytorch_model-00018-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cd738bc716430aaf7425eeb7762001842b2b7e9a195a6f9b39ea4b506bb1ee3b
+size 4857113187

pytorch_model-00019-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e84ba2f0efc901548aec67c4b36a9bb7dc770c1f52f6431fb4813135d7a5e495
+size 4857113187

pytorch_model-00020-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:53fae31b8578a38c5ef74fc55f6a48608f4f7b41c270abdb5ecbcec25dad6758
+size 4857113187

pytorch_model-00021-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a53f12d2f9564a9965b98688265878a8a2fead55c80fc722f7ee03bb9e15d574
+size 4857113187

pytorch_model-00022-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ada24e8b5c959e6aa73a1f44318e1bc47c14ac09a0e1b67c72c98b67da0e53e5
+size 4857113187

pytorch_model-00023-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:148bca292f74ae8b924d7826205362e4979b8b39c9f6a87a7d355ba67ca1e948
+size 4857113187

pytorch_model-00024-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a20dbcfd5b0b7048be56cb75932a3cbccfff189ff58a24a80f33431dfc9af8a
+size 4857113187

pytorch_model-00025-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a3c4b19b86b3e41eebbc68c5cf4341135b0375ecbf44f2811940ab9498ce8e6
+size 4857113187

pytorch_model-00026-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc7d7be5810786007ebdef16fb110ad817b0d58a6ebbdcd1e69196b857eaafef
+size 4857113187

pytorch_model-00027-of-00027.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b182b4625f41bd805de99744b1e675c3f7f99b3c3757e0b9b96b97ab543e221b
+size 4483834935

pytorch_model.bin.index.json

@@ -0,0 +1,810 @@
+{
+  "metadata": {
+    "total_size": 130571341824
+  },
+  "weight_map": {
+    "lm_head.weight": "pytorch_model-00027-of-00027.bin",
+    "model.embed_tokens.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.0.input_layernorm.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.0.mlp.down_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.0.mlp.gate_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.0.mlp.up_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.0.post_attention_layernorm.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.0.self_attn.k_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.0.self_attn.o_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.0.self_attn.q_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.0.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00027.bin",
+    "model.layers.0.self_attn.v_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.1.input_layernorm.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.1.mlp.down_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.1.mlp.gate_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.1.mlp.up_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.1.post_attention_layernorm.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.1.self_attn.k_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.1.self_attn.o_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.1.self_attn.q_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.1.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00027.bin",
+    "model.layers.1.self_attn.v_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.10.input_layernorm.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.10.mlp.down_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.10.mlp.gate_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.10.mlp.up_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.10.post_attention_layernorm.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.10.self_attn.k_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.10.self_attn.o_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.10.self_attn.q_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.10.self_attn.rotary_emb.inv_freq": "pytorch_model-00004-of-00027.bin",
+    "model.layers.10.self_attn.v_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.11.input_layernorm.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.11.mlp.down_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.11.mlp.gate_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.11.mlp.up_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.11.post_attention_layernorm.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.11.self_attn.k_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.11.self_attn.o_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.11.self_attn.q_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.11.self_attn.rotary_emb.inv_freq": "pytorch_model-00004-of-00027.bin",
+    "model.layers.11.self_attn.v_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.12.input_layernorm.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.12.mlp.down_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.12.mlp.gate_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.12.mlp.up_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.12.post_attention_layernorm.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.12.self_attn.k_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.12.self_attn.o_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.12.self_attn.q_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.12.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00027.bin",
+    "model.layers.12.self_attn.v_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.13.input_layernorm.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.13.mlp.down_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.13.mlp.gate_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.13.mlp.up_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.13.post_attention_layernorm.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.13.self_attn.k_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.13.self_attn.o_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.13.self_attn.q_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.13.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00027.bin",
+    "model.layers.13.self_attn.v_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.14.input_layernorm.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.14.mlp.down_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.14.mlp.gate_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.14.mlp.up_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.14.post_attention_layernorm.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.14.self_attn.k_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.14.self_attn.o_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.14.self_attn.q_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.14.self_attn.rotary_emb.inv_freq": "pytorch_model-00005-of-00027.bin",
+    "model.layers.14.self_attn.v_proj.weight": "pytorch_model-00005-of-00027.bin",
+    "model.layers.15.input_layernorm.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.15.mlp.down_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.15.mlp.gate_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.15.mlp.up_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.15.post_attention_layernorm.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.15.self_attn.k_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.15.self_attn.o_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.15.self_attn.q_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.15.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00027.bin",
+    "model.layers.15.self_attn.v_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.16.input_layernorm.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.16.mlp.down_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.16.mlp.gate_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.16.mlp.up_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.16.post_attention_layernorm.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.16.self_attn.k_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.16.self_attn.o_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.16.self_attn.q_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.16.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00027.bin",
+    "model.layers.16.self_attn.v_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.17.input_layernorm.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.17.mlp.down_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.17.mlp.gate_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.17.mlp.up_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.17.post_attention_layernorm.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.17.self_attn.k_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.17.self_attn.o_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.17.self_attn.q_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.17.self_attn.rotary_emb.inv_freq": "pytorch_model-00006-of-00027.bin",
+    "model.layers.17.self_attn.v_proj.weight": "pytorch_model-00006-of-00027.bin",
+    "model.layers.18.input_layernorm.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.18.mlp.down_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.18.mlp.gate_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.18.mlp.up_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.18.post_attention_layernorm.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.18.self_attn.k_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.18.self_attn.o_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.18.self_attn.q_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.18.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00027.bin",
+    "model.layers.18.self_attn.v_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.19.input_layernorm.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.19.mlp.down_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.19.mlp.gate_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.19.mlp.up_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.19.post_attention_layernorm.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.19.self_attn.k_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.19.self_attn.o_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.19.self_attn.q_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.19.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00027.bin",
+    "model.layers.19.self_attn.v_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.2.input_layernorm.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.2.mlp.down_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.2.mlp.gate_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.2.mlp.up_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.2.post_attention_layernorm.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.2.self_attn.k_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.2.self_attn.o_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.2.self_attn.q_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.2.self_attn.rotary_emb.inv_freq": "pytorch_model-00001-of-00027.bin",
+    "model.layers.2.self_attn.v_proj.weight": "pytorch_model-00001-of-00027.bin",
+    "model.layers.20.input_layernorm.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.20.mlp.down_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.20.mlp.gate_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.20.mlp.up_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.20.post_attention_layernorm.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.20.self_attn.k_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.20.self_attn.o_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.20.self_attn.q_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.20.self_attn.rotary_emb.inv_freq": "pytorch_model-00007-of-00027.bin",
+    "model.layers.20.self_attn.v_proj.weight": "pytorch_model-00007-of-00027.bin",
+    "model.layers.21.input_layernorm.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.21.mlp.down_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.21.mlp.gate_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.21.mlp.up_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.21.post_attention_layernorm.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.21.self_attn.k_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.21.self_attn.o_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.21.self_attn.q_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.21.self_attn.rotary_emb.inv_freq": "pytorch_model-00008-of-00027.bin",
+    "model.layers.21.self_attn.v_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.22.input_layernorm.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.22.mlp.down_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.22.mlp.gate_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.22.mlp.up_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.22.post_attention_layernorm.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.22.self_attn.k_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.22.self_attn.o_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.22.self_attn.q_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.22.self_attn.rotary_emb.inv_freq": "pytorch_model-00008-of-00027.bin",
+    "model.layers.22.self_attn.v_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.23.input_layernorm.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.23.mlp.down_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.23.mlp.gate_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.23.mlp.up_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.23.post_attention_layernorm.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.23.self_attn.k_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.23.self_attn.o_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.23.self_attn.q_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.23.self_attn.rotary_emb.inv_freq": "pytorch_model-00008-of-00027.bin",
+    "model.layers.23.self_attn.v_proj.weight": "pytorch_model-00008-of-00027.bin",
+    "model.layers.24.input_layernorm.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.24.mlp.down_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.24.mlp.gate_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.24.mlp.up_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.24.post_attention_layernorm.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.24.self_attn.k_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.24.self_attn.o_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.24.self_attn.q_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.24.self_attn.rotary_emb.inv_freq": "pytorch_model-00009-of-00027.bin",
+    "model.layers.24.self_attn.v_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.25.input_layernorm.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.25.mlp.down_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.25.mlp.gate_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.25.mlp.up_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.25.post_attention_layernorm.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.25.self_attn.k_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.25.self_attn.o_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.25.self_attn.q_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.25.self_attn.rotary_emb.inv_freq": "pytorch_model-00009-of-00027.bin",
+    "model.layers.25.self_attn.v_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.26.input_layernorm.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.26.mlp.down_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.26.mlp.gate_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.26.mlp.up_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.26.post_attention_layernorm.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.26.self_attn.k_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.26.self_attn.o_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.26.self_attn.q_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.26.self_attn.rotary_emb.inv_freq": "pytorch_model-00009-of-00027.bin",
+    "model.layers.26.self_attn.v_proj.weight": "pytorch_model-00009-of-00027.bin",
+    "model.layers.27.input_layernorm.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.27.mlp.down_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.27.mlp.gate_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.27.mlp.up_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.27.post_attention_layernorm.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.27.self_attn.k_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.27.self_attn.o_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.27.self_attn.q_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.27.self_attn.rotary_emb.inv_freq": "pytorch_model-00010-of-00027.bin",
+    "model.layers.27.self_attn.v_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.28.input_layernorm.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.28.mlp.down_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.28.mlp.gate_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.28.mlp.up_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.28.post_attention_layernorm.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.28.self_attn.k_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.28.self_attn.o_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.28.self_attn.q_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.28.self_attn.rotary_emb.inv_freq": "pytorch_model-00010-of-00027.bin",
+    "model.layers.28.self_attn.v_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.29.input_layernorm.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.29.mlp.down_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.29.mlp.gate_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.29.mlp.up_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.29.post_attention_layernorm.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.29.self_attn.k_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.29.self_attn.o_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.29.self_attn.q_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.29.self_attn.rotary_emb.inv_freq": "pytorch_model-00010-of-00027.bin",
+    "model.layers.29.self_attn.v_proj.weight": "pytorch_model-00010-of-00027.bin",
+    "model.layers.3.input_layernorm.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.3.mlp.down_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.3.mlp.gate_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.3.mlp.up_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.3.post_attention_layernorm.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.3.self_attn.k_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.3.self_attn.o_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.3.self_attn.q_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.3.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00027.bin",
+    "model.layers.3.self_attn.v_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.30.input_layernorm.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.30.mlp.down_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.30.mlp.gate_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.30.mlp.up_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.30.post_attention_layernorm.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.30.self_attn.k_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.30.self_attn.o_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.30.self_attn.q_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.30.self_attn.rotary_emb.inv_freq": "pytorch_model-00011-of-00027.bin",
+    "model.layers.30.self_attn.v_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.31.input_layernorm.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.31.mlp.down_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.31.mlp.gate_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.31.mlp.up_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.31.post_attention_layernorm.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.31.self_attn.k_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.31.self_attn.o_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.31.self_attn.q_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.31.self_attn.rotary_emb.inv_freq": "pytorch_model-00011-of-00027.bin",
+    "model.layers.31.self_attn.v_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.32.input_layernorm.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.32.mlp.down_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.32.mlp.gate_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.32.mlp.up_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.32.post_attention_layernorm.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.32.self_attn.k_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.32.self_attn.o_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.32.self_attn.q_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.32.self_attn.rotary_emb.inv_freq": "pytorch_model-00011-of-00027.bin",
+    "model.layers.32.self_attn.v_proj.weight": "pytorch_model-00011-of-00027.bin",
+    "model.layers.33.input_layernorm.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.33.mlp.down_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.33.mlp.gate_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.33.mlp.up_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.33.post_attention_layernorm.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.33.self_attn.k_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.33.self_attn.o_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.33.self_attn.q_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.33.self_attn.rotary_emb.inv_freq": "pytorch_model-00012-of-00027.bin",
+    "model.layers.33.self_attn.v_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.34.input_layernorm.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.34.mlp.down_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.34.mlp.gate_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.34.mlp.up_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.34.post_attention_layernorm.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.34.self_attn.k_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.34.self_attn.o_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.34.self_attn.q_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.34.self_attn.rotary_emb.inv_freq": "pytorch_model-00012-of-00027.bin",
+    "model.layers.34.self_attn.v_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.35.input_layernorm.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.35.mlp.down_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.35.mlp.gate_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.35.mlp.up_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.35.post_attention_layernorm.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.35.self_attn.k_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.35.self_attn.o_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.35.self_attn.q_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.35.self_attn.rotary_emb.inv_freq": "pytorch_model-00012-of-00027.bin",
+    "model.layers.35.self_attn.v_proj.weight": "pytorch_model-00012-of-00027.bin",
+    "model.layers.36.input_layernorm.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.36.mlp.down_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.36.mlp.gate_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.36.mlp.up_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.36.post_attention_layernorm.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.36.self_attn.k_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.36.self_attn.o_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.36.self_attn.q_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.36.self_attn.rotary_emb.inv_freq": "pytorch_model-00013-of-00027.bin",
+    "model.layers.36.self_attn.v_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.37.input_layernorm.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.37.mlp.down_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.37.mlp.gate_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.37.mlp.up_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.37.post_attention_layernorm.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.37.self_attn.k_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.37.self_attn.o_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.37.self_attn.q_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.37.self_attn.rotary_emb.inv_freq": "pytorch_model-00013-of-00027.bin",
+    "model.layers.37.self_attn.v_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.38.input_layernorm.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.38.mlp.down_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.38.mlp.gate_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.38.mlp.up_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.38.post_attention_layernorm.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.38.self_attn.k_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.38.self_attn.o_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.38.self_attn.q_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.38.self_attn.rotary_emb.inv_freq": "pytorch_model-00013-of-00027.bin",
+    "model.layers.38.self_attn.v_proj.weight": "pytorch_model-00013-of-00027.bin",
+    "model.layers.39.input_layernorm.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.39.mlp.down_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.39.mlp.gate_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.39.mlp.up_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.39.post_attention_layernorm.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.39.self_attn.k_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.39.self_attn.o_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.39.self_attn.q_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.39.self_attn.rotary_emb.inv_freq": "pytorch_model-00014-of-00027.bin",
+    "model.layers.39.self_attn.v_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.4.input_layernorm.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.4.mlp.down_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.4.mlp.gate_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.4.mlp.up_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.4.post_attention_layernorm.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.4.self_attn.k_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.4.self_attn.o_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.4.self_attn.q_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.4.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00027.bin",
+    "model.layers.4.self_attn.v_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.40.input_layernorm.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.40.mlp.down_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.40.mlp.gate_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.40.mlp.up_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.40.post_attention_layernorm.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.40.self_attn.k_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.40.self_attn.o_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.40.self_attn.q_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.40.self_attn.rotary_emb.inv_freq": "pytorch_model-00014-of-00027.bin",
+    "model.layers.40.self_attn.v_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.41.input_layernorm.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.41.mlp.down_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.41.mlp.gate_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.41.mlp.up_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.41.post_attention_layernorm.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.41.self_attn.k_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.41.self_attn.o_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.41.self_attn.q_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.41.self_attn.rotary_emb.inv_freq": "pytorch_model-00014-of-00027.bin",
+    "model.layers.41.self_attn.v_proj.weight": "pytorch_model-00014-of-00027.bin",
+    "model.layers.42.input_layernorm.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.42.mlp.down_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.42.mlp.gate_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.42.mlp.up_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.42.post_attention_layernorm.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.42.self_attn.k_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.42.self_attn.o_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.42.self_attn.q_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.42.self_attn.rotary_emb.inv_freq": "pytorch_model-00015-of-00027.bin",
+    "model.layers.42.self_attn.v_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.43.input_layernorm.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.43.mlp.down_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.43.mlp.gate_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.43.mlp.up_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.43.post_attention_layernorm.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.43.self_attn.k_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.43.self_attn.o_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.43.self_attn.q_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.43.self_attn.rotary_emb.inv_freq": "pytorch_model-00015-of-00027.bin",
+    "model.layers.43.self_attn.v_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.44.input_layernorm.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.44.mlp.down_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.44.mlp.gate_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.44.mlp.up_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.44.post_attention_layernorm.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.44.self_attn.k_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.44.self_attn.o_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.44.self_attn.q_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.44.self_attn.rotary_emb.inv_freq": "pytorch_model-00015-of-00027.bin",
+    "model.layers.44.self_attn.v_proj.weight": "pytorch_model-00015-of-00027.bin",
+    "model.layers.45.input_layernorm.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.45.mlp.down_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.45.mlp.gate_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.45.mlp.up_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.45.post_attention_layernorm.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.45.self_attn.k_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.45.self_attn.o_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.45.self_attn.q_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.45.self_attn.rotary_emb.inv_freq": "pytorch_model-00016-of-00027.bin",
+    "model.layers.45.self_attn.v_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.46.input_layernorm.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.46.mlp.down_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.46.mlp.gate_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.46.mlp.up_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.46.post_attention_layernorm.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.46.self_attn.k_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.46.self_attn.o_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.46.self_attn.q_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.46.self_attn.rotary_emb.inv_freq": "pytorch_model-00016-of-00027.bin",
+    "model.layers.46.self_attn.v_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.47.input_layernorm.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.47.mlp.down_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.47.mlp.gate_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.47.mlp.up_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.47.post_attention_layernorm.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.47.self_attn.k_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.47.self_attn.o_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.47.self_attn.q_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.47.self_attn.rotary_emb.inv_freq": "pytorch_model-00016-of-00027.bin",
+    "model.layers.47.self_attn.v_proj.weight": "pytorch_model-00016-of-00027.bin",
+    "model.layers.48.input_layernorm.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.48.mlp.down_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.48.mlp.gate_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.48.mlp.up_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.48.post_attention_layernorm.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.48.self_attn.k_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.48.self_attn.o_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.48.self_attn.q_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.48.self_attn.rotary_emb.inv_freq": "pytorch_model-00017-of-00027.bin",
+    "model.layers.48.self_attn.v_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.49.input_layernorm.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.49.mlp.down_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.49.mlp.gate_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.49.mlp.up_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.49.post_attention_layernorm.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.49.self_attn.k_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.49.self_attn.o_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.49.self_attn.q_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.49.self_attn.rotary_emb.inv_freq": "pytorch_model-00017-of-00027.bin",
+    "model.layers.49.self_attn.v_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.5.input_layernorm.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.5.mlp.down_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.5.mlp.gate_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.5.mlp.up_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.5.post_attention_layernorm.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.5.self_attn.k_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.5.self_attn.o_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.5.self_attn.q_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.5.self_attn.rotary_emb.inv_freq": "pytorch_model-00002-of-00027.bin",
+    "model.layers.5.self_attn.v_proj.weight": "pytorch_model-00002-of-00027.bin",
+    "model.layers.50.input_layernorm.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.50.mlp.down_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.50.mlp.gate_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.50.mlp.up_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.50.post_attention_layernorm.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.50.self_attn.k_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.50.self_attn.o_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.50.self_attn.q_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.50.self_attn.rotary_emb.inv_freq": "pytorch_model-00017-of-00027.bin",
+    "model.layers.50.self_attn.v_proj.weight": "pytorch_model-00017-of-00027.bin",
+    "model.layers.51.input_layernorm.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.51.mlp.down_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.51.mlp.gate_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.51.mlp.up_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.51.post_attention_layernorm.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.51.self_attn.k_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.51.self_attn.o_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.51.self_attn.q_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.51.self_attn.rotary_emb.inv_freq": "pytorch_model-00018-of-00027.bin",
+    "model.layers.51.self_attn.v_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.52.input_layernorm.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.52.mlp.down_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.52.mlp.gate_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.52.mlp.up_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.52.post_attention_layernorm.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.52.self_attn.k_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.52.self_attn.o_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.52.self_attn.q_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.52.self_attn.rotary_emb.inv_freq": "pytorch_model-00018-of-00027.bin",
+    "model.layers.52.self_attn.v_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.53.input_layernorm.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.53.mlp.down_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.53.mlp.gate_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.53.mlp.up_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.53.post_attention_layernorm.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.53.self_attn.k_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.53.self_attn.o_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.53.self_attn.q_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.53.self_attn.rotary_emb.inv_freq": "pytorch_model-00018-of-00027.bin",
+    "model.layers.53.self_attn.v_proj.weight": "pytorch_model-00018-of-00027.bin",
+    "model.layers.54.input_layernorm.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.54.mlp.down_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.54.mlp.gate_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.54.mlp.up_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.54.post_attention_layernorm.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.54.self_attn.k_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.54.self_attn.o_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.54.self_attn.q_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.54.self_attn.rotary_emb.inv_freq": "pytorch_model-00019-of-00027.bin",
+    "model.layers.54.self_attn.v_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.55.input_layernorm.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.55.mlp.down_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.55.mlp.gate_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.55.mlp.up_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.55.post_attention_layernorm.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.55.self_attn.k_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.55.self_attn.o_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.55.self_attn.q_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.55.self_attn.rotary_emb.inv_freq": "pytorch_model-00019-of-00027.bin",
+    "model.layers.55.self_attn.v_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.56.input_layernorm.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.56.mlp.down_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.56.mlp.gate_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.56.mlp.up_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.56.post_attention_layernorm.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.56.self_attn.k_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.56.self_attn.o_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.56.self_attn.q_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.56.self_attn.rotary_emb.inv_freq": "pytorch_model-00019-of-00027.bin",
+    "model.layers.56.self_attn.v_proj.weight": "pytorch_model-00019-of-00027.bin",
+    "model.layers.57.input_layernorm.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.57.mlp.down_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.57.mlp.gate_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.57.mlp.up_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.57.post_attention_layernorm.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.57.self_attn.k_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.57.self_attn.o_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.57.self_attn.q_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.57.self_attn.rotary_emb.inv_freq": "pytorch_model-00020-of-00027.bin",
+    "model.layers.57.self_attn.v_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.58.input_layernorm.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.58.mlp.down_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.58.mlp.gate_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.58.mlp.up_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.58.post_attention_layernorm.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.58.self_attn.k_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.58.self_attn.o_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.58.self_attn.q_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.58.self_attn.rotary_emb.inv_freq": "pytorch_model-00020-of-00027.bin",
+    "model.layers.58.self_attn.v_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.59.input_layernorm.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.59.mlp.down_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.59.mlp.gate_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.59.mlp.up_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.59.post_attention_layernorm.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.59.self_attn.k_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.59.self_attn.o_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.59.self_attn.q_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.59.self_attn.rotary_emb.inv_freq": "pytorch_model-00020-of-00027.bin",
+    "model.layers.59.self_attn.v_proj.weight": "pytorch_model-00020-of-00027.bin",
+    "model.layers.6.input_layernorm.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.6.mlp.down_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.6.mlp.gate_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.6.mlp.up_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.6.post_attention_layernorm.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.6.self_attn.k_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.6.self_attn.o_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.6.self_attn.q_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.6.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00027.bin",
+    "model.layers.6.self_attn.v_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.60.input_layernorm.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.60.mlp.down_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.60.mlp.gate_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.60.mlp.up_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.60.post_attention_layernorm.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.60.self_attn.k_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.60.self_attn.o_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.60.self_attn.q_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.60.self_attn.rotary_emb.inv_freq": "pytorch_model-00021-of-00027.bin",
+    "model.layers.60.self_attn.v_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.61.input_layernorm.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.61.mlp.down_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.61.mlp.gate_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.61.mlp.up_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.61.post_attention_layernorm.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.61.self_attn.k_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.61.self_attn.o_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.61.self_attn.q_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.61.self_attn.rotary_emb.inv_freq": "pytorch_model-00021-of-00027.bin",
+    "model.layers.61.self_attn.v_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.62.input_layernorm.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.62.mlp.down_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.62.mlp.gate_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.62.mlp.up_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.62.post_attention_layernorm.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.62.self_attn.k_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.62.self_attn.o_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.62.self_attn.q_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.62.self_attn.rotary_emb.inv_freq": "pytorch_model-00021-of-00027.bin",
+    "model.layers.62.self_attn.v_proj.weight": "pytorch_model-00021-of-00027.bin",
+    "model.layers.63.input_layernorm.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.63.mlp.down_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.63.mlp.gate_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.63.mlp.up_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.63.post_attention_layernorm.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.63.self_attn.k_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.63.self_attn.o_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.63.self_attn.q_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.63.self_attn.rotary_emb.inv_freq": "pytorch_model-00022-of-00027.bin",
+    "model.layers.63.self_attn.v_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.64.input_layernorm.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.64.mlp.down_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.64.mlp.gate_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.64.mlp.up_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.64.post_attention_layernorm.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.64.self_attn.k_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.64.self_attn.o_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.64.self_attn.q_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.64.self_attn.rotary_emb.inv_freq": "pytorch_model-00022-of-00027.bin",
+    "model.layers.64.self_attn.v_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.65.input_layernorm.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.65.mlp.down_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.65.mlp.gate_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.65.mlp.up_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.65.post_attention_layernorm.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.65.self_attn.k_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.65.self_attn.o_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.65.self_attn.q_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.65.self_attn.rotary_emb.inv_freq": "pytorch_model-00022-of-00027.bin",
+    "model.layers.65.self_attn.v_proj.weight": "pytorch_model-00022-of-00027.bin",
+    "model.layers.66.input_layernorm.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.66.mlp.down_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.66.mlp.gate_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.66.mlp.up_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.66.post_attention_layernorm.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.66.self_attn.k_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.66.self_attn.o_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.66.self_attn.q_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.66.self_attn.rotary_emb.inv_freq": "pytorch_model-00023-of-00027.bin",
+    "model.layers.66.self_attn.v_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.67.input_layernorm.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.67.mlp.down_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.67.mlp.gate_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.67.mlp.up_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.67.post_attention_layernorm.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.67.self_attn.k_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.67.self_attn.o_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.67.self_attn.q_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.67.self_attn.rotary_emb.inv_freq": "pytorch_model-00023-of-00027.bin",
+    "model.layers.67.self_attn.v_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.68.input_layernorm.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.68.mlp.down_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.68.mlp.gate_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.68.mlp.up_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.68.post_attention_layernorm.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.68.self_attn.k_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.68.self_attn.o_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.68.self_attn.q_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.68.self_attn.rotary_emb.inv_freq": "pytorch_model-00023-of-00027.bin",
+    "model.layers.68.self_attn.v_proj.weight": "pytorch_model-00023-of-00027.bin",
+    "model.layers.69.input_layernorm.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.69.mlp.down_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.69.mlp.gate_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.69.mlp.up_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.69.post_attention_layernorm.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.69.self_attn.k_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.69.self_attn.o_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.69.self_attn.q_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.69.self_attn.rotary_emb.inv_freq": "pytorch_model-00024-of-00027.bin",
+    "model.layers.69.self_attn.v_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.7.input_layernorm.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.7.mlp.down_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.7.mlp.gate_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.7.mlp.up_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.7.post_attention_layernorm.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.7.self_attn.k_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.7.self_attn.o_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.7.self_attn.q_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.7.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00027.bin",
+    "model.layers.7.self_attn.v_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.70.input_layernorm.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.70.mlp.down_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.70.mlp.gate_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.70.mlp.up_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.70.post_attention_layernorm.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.70.self_attn.k_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.70.self_attn.o_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.70.self_attn.q_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.70.self_attn.rotary_emb.inv_freq": "pytorch_model-00024-of-00027.bin",
+    "model.layers.70.self_attn.v_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.71.input_layernorm.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.71.mlp.down_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.71.mlp.gate_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.71.mlp.up_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.71.post_attention_layernorm.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.71.self_attn.k_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.71.self_attn.o_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.71.self_attn.q_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.71.self_attn.rotary_emb.inv_freq": "pytorch_model-00024-of-00027.bin",
+    "model.layers.71.self_attn.v_proj.weight": "pytorch_model-00024-of-00027.bin",
+    "model.layers.72.input_layernorm.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.72.mlp.down_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.72.mlp.gate_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.72.mlp.up_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.72.post_attention_layernorm.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.72.self_attn.k_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.72.self_attn.o_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.72.self_attn.q_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.72.self_attn.rotary_emb.inv_freq": "pytorch_model-00025-of-00027.bin",
+    "model.layers.72.self_attn.v_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.73.input_layernorm.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.73.mlp.down_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.73.mlp.gate_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.73.mlp.up_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.73.post_attention_layernorm.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.73.self_attn.k_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.73.self_attn.o_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.73.self_attn.q_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.73.self_attn.rotary_emb.inv_freq": "pytorch_model-00025-of-00027.bin",
+    "model.layers.73.self_attn.v_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.74.input_layernorm.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.74.mlp.down_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.74.mlp.gate_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.74.mlp.up_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.74.post_attention_layernorm.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.74.self_attn.k_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.74.self_attn.o_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.74.self_attn.q_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.74.self_attn.rotary_emb.inv_freq": "pytorch_model-00025-of-00027.bin",
+    "model.layers.74.self_attn.v_proj.weight": "pytorch_model-00025-of-00027.bin",
+    "model.layers.75.input_layernorm.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.75.mlp.down_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.75.mlp.gate_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.75.mlp.up_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.75.post_attention_layernorm.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.75.self_attn.k_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.75.self_attn.o_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.75.self_attn.q_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.75.self_attn.rotary_emb.inv_freq": "pytorch_model-00026-of-00027.bin",
+    "model.layers.75.self_attn.v_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.76.input_layernorm.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.76.mlp.down_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.76.mlp.gate_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.76.mlp.up_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.76.post_attention_layernorm.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.76.self_attn.k_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.76.self_attn.o_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.76.self_attn.q_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.76.self_attn.rotary_emb.inv_freq": "pytorch_model-00026-of-00027.bin",
+    "model.layers.76.self_attn.v_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.77.input_layernorm.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.77.mlp.down_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.77.mlp.gate_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.77.mlp.up_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.77.post_attention_layernorm.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.77.self_attn.k_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.77.self_attn.o_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.77.self_attn.q_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.77.self_attn.rotary_emb.inv_freq": "pytorch_model-00026-of-00027.bin",
+    "model.layers.77.self_attn.v_proj.weight": "pytorch_model-00026-of-00027.bin",
+    "model.layers.78.input_layernorm.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.78.mlp.down_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.78.mlp.gate_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.78.mlp.up_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.78.post_attention_layernorm.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.78.self_attn.k_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.78.self_attn.o_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.78.self_attn.q_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.78.self_attn.rotary_emb.inv_freq": "pytorch_model-00027-of-00027.bin",
+    "model.layers.78.self_attn.v_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.79.input_layernorm.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.79.mlp.down_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.79.mlp.gate_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.79.mlp.up_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.79.post_attention_layernorm.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.79.self_attn.k_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.79.self_attn.o_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.79.self_attn.q_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.79.self_attn.rotary_emb.inv_freq": "pytorch_model-00027-of-00027.bin",
+    "model.layers.79.self_attn.v_proj.weight": "pytorch_model-00027-of-00027.bin",
+    "model.layers.8.input_layernorm.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.8.mlp.down_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.8.mlp.gate_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.8.mlp.up_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.8.post_attention_layernorm.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.8.self_attn.k_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.8.self_attn.o_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.8.self_attn.q_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.8.self_attn.rotary_emb.inv_freq": "pytorch_model-00003-of-00027.bin",
+    "model.layers.8.self_attn.v_proj.weight": "pytorch_model-00003-of-00027.bin",
+    "model.layers.9.input_layernorm.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.9.mlp.down_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.9.mlp.gate_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.9.mlp.up_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.9.post_attention_layernorm.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.9.self_attn.k_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.9.self_attn.o_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.9.self_attn.q_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.layers.9.self_attn.rotary_emb.inv_freq": "pytorch_model-00004-of-00027.bin",
+    "model.layers.9.self_attn.v_proj.weight": "pytorch_model-00004-of-00027.bin",
+    "model.norm.weight": "pytorch_model-00027-of-00027.bin"
+  }
+}

special_tokens_map.json

@@ -0,0 +1,23 @@
+{
+  "bos_token": {
+    "content": "",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723

tokenizer_config.json

@@ -0,0 +1,33 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "bos_token": {
+    "__type": "AddedToken",
+    "content": "",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "clean_up_tokenization_spaces": false,
+  "eos_token": {
+    "__type": "AddedToken",
+    "content": "",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  },
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": null,
+  "sp_model_kwargs": {},
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": {
+    "__type": "AddedToken",
+    "content": "",
+    "lstrip": false,
+    "normalized": true,
+    "rstrip": false,
+    "single_word": false
+  }
+}