Merge branch 'main' into refactor/rename-4b-to-gptq

README.md

@@ -14,12 +14,13 @@
 
 ## Axolotl supports
 
-|          | fp16/fp32 | fp16/fp32 w/ lora | 4bit-quant | 4bit-quant w/flash attention | flash attention | xformers attention |
-|----------|:----------|:------------------|------------|------------------------------|-----------------|--------------------|
-| llama    | ✅         | ✅                 | ✅          | ✅                            | ✅               | ✅                  |
-| Pythia   | ✅         | ✅                 | ❌          | ❌                            | ❌               | ❓                  |
-| cerebras | ✅         | ✅                 | ❌          | ❌                            | ❌               | ❓                  |
-| mpt      | ✅         | ❌                 | ❌          | ❌                            | ❌               | ❓                  |
+|         | fp16/fp32 | fp16/fp32 w/ lora | qlora | 4bit-quant | 4bit-quant w/flash attention | flash attention | xformers attention |
+|---------|:----------|:------------------|------|------------|------------------------------|-----------------|--------------------|
+| llama   | ✅         | ✅                 | ✅  | ✅          | ✅                            | ✅               | ✅                  |
+| Pythia  | ✅         | ✅                 | ❓  | ❌          | ❌                            | ❌               | ❓                  |
+| cerebras | ✅         | ✅                 | ❓  | ❌          | ❌                            | ❌               | ❓                  |
+| mpt     | ✅         | ❌                 | ❓  | ❌          | ❌                            | ❌               | ❓                  |
+| falcon  | ✅         | ❌                 | ❌  | ❌          | ❌                            | ❌               | ❓                  |
 
 
 ## Quickstart ⚡

examples/falcon/config-7b-lora.yml

@@ -0,0 +1,64 @@
+base_model: tiiuae/falcon-7b
+base_model_config: tiiuae/falcon-7b
+trust_remote_code: true
+model_type: AutoModelForCausalLM
+tokenizer_type: AutoTokenizer
+load_in_8bit: true
+load_in_4bit: false
+gptq: false
+strict: false
+push_dataset_to_hub:
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca:chat
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+adapter: lora
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len:
+lora_r: 16
+lora_alpha: 32
+lora_dropout: 0.0
+lora_target_modules:
+lora_target_linear: true
+lora_fan_in_fan_out:
+wandb_project: falcon-7b
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./falcon-7b
+batch_size: 2
+micro_batch_size: 1
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.00003
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: true
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 40
+eval_steps: 5
+save_steps: 43
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:
+  pad_token: "<|endoftext|>"
+  bos_token: ">>ABSTRACT<<"
+  eos_token: "<|endoftext|>"
+

examples/falcon/config-7b.yml

@@ -0,0 +1,64 @@
+base_model: tiiuae/falcon-7b
+base_model_config: tiiuae/falcon-7b
+trust_remote_code: true
+model_type: AutoModelForCausalLM
+tokenizer_type: AutoTokenizer
+load_in_8bit: false
+load_in_4bit: false
+gptq: false
+strict: false
+push_dataset_to_hub:
+datasets:
+  - path: teknium/GPT4-LLM-Cleaned
+    type: alpaca:chat
+dataset_prepared_path: last_run_prepared
+val_set_size: 0.01
+adapter:
+lora_model_dir:
+sequence_len: 2048
+max_packed_sequence_len:
+lora_r: 64
+lora_alpha: 32
+lora_dropout: 0.0
+lora_target_modules:
+lora_target_linear: true
+lora_fan_in_fan_out:
+wandb_project: falcon-7b
+wandb_watch:
+wandb_run_id:
+wandb_log_model:
+output_dir: ./falcon-7b
+batch_size: 2
+micro_batch_size: 1
+num_epochs: 4
+optimizer: adamw_bnb_8bit
+torchdistx_path:
+lr_scheduler: cosine
+learning_rate: 0.00003
+train_on_inputs: false
+group_by_length: false
+bf16: true
+fp16: false
+tf32: true
+gradient_checkpointing: true
+early_stopping_patience:
+resume_from_checkpoint:
+local_rank:
+logging_steps: 1
+xformers_attention: true
+flash_attention:
+gptq_groupsize:
+gptq_model_v1:
+warmup_steps: 40
+eval_steps: 5
+save_steps: 43
+debug:
+deepspeed:
+weight_decay: 0.0
+fsdp:
+fsdp_config:
+special_tokens:
+  pad_token: "<|endoftext|>"
+  bos_token: ">>ABSTRACT<<"
+  eos_token: "<|endoftext|>"
+

src/axolotl/utils/models.py

@@ -364,7 +364,7 @@ def load_lora(model, cfg):
         PeftModel,
     )
 
-    lora_target_modules = list(cfg.lora_target_modules)
+    lora_target_modules = list(cfg.lora_target_modules or [])
 
     if cfg.lora_target_linear:
         bits = None

src/axolotl/utils/validation.py

@@ -8,12 +8,12 @@ def validate_config(cfg):
     if cfg.adapter == "qlora":
         if cfg.merge_lora:
             # can't merge qlora if loaded in 8bit or 4bit
-            assert cfg.load_in_8bit is False
-            assert cfg.gptq is False
-            assert cfg.load_in_4bit is False
+            assert cfg.load_in_8bit is not True
+            assert cfg.gptq is not True
+            assert cfg.load_in_4bit is not True
         else:
-            assert cfg.load_in_8bit is False
-            assert cfg.gptq is False
+            assert cfg.load_in_8bit is not True
+            assert cfg.gptq is not True
             assert cfg.load_in_4bit is True
 
     if not cfg.load_in_8bit and cfg.adapter == "lora":