skip s2 attention test due to timeout

DoRA with quantized layers is supported with PEFT 0.10.0

src/axolotl/cli/merge_lora.py

@@ -38,6 +38,8 @@ def do_cli(config: Path = Path("examples/"), **kwargs):
     parsed_cfg.load_in_4bit = False
     parsed_cfg.load_in_8bit = False
     parsed_cfg.flash_attention = False
+    parsed_cfg.deepspeed = None
+    parsed_cfg.fsdp = None
 
     do_merge_lora(cfg=parsed_cfg, cli_args=parsed_cli_args)
 

src/axolotl/core/trainer_builder.py

@@ -23,6 +23,7 @@ from torch.optim.lr_scheduler import OneCycleLR
 from torch.utils.data import BatchSampler, DataLoader, RandomSampler, SequentialSampler
 from transformers import (
     EarlyStoppingCallback,
+    PreTrainedModel,
     Trainer,
     TrainerCallback,
     TrainingArguments,
@@ -802,6 +803,15 @@ class AxolotlDPOTrainer(DPOTrainer):
 
         return super().push_to_hub(*args, **kwargs)
 
+    def tokenize_row(
+        self, feature, model: Optional[Union[PreTrainedModel, torch.nn.Module]] = None
+    ) -> Dict:
+        res = super().tokenize_row(feature, model=model)
+        if self.tokenizer.bos_token_id is None and res["prompt_input_ids"][0] is None:
+            for key in res.keys():
+                res[key] = res[key][1:]
+        return res
+
 
 class TrainerBuilderBase(abc.ABC):
     """

src/axolotl/utils/chat_templates.py

@@ -23,6 +23,7 @@ def chat_templates(user_choice: str):
         "inst": "{{ bos_token }}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if message['role'] == 'user' %}{{ '[INST] ' + message['content'] + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ message['content'] + eos_token}}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}",  # I don't know what this one is called. Used by Mistral/Mixtral.
         "chatml": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
         "gemma": "{{ bos_token }}{% if messages[0]['role'] == 'system' %}{{ raise_exception('System role not supported') }}{% endif %}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if (message['role'] == 'assistant') %}{% set role = 'model' %}{% else %}{% set role = message['role'] %}{% endif %}{{ '<start_of_turn>' + role + '\n' + message['content'] | trim + '<end_of_turn>\n' }}{% endfor %}{% if add_generation_prompt %}{{'<start_of_turn>model\n'}}{% endif %}",
+        "cohere": "{{ bos_token }}{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% elif false == true %}{% set loop_messages = messages %}{% set system_message = 'You are Command-R, a brilliant, sophisticated, AI-assistant trained to assist human users by providing thorough responses. You are trained by Cohere.' %}{% else %}{% set loop_messages = messages %}{% set system_message = false %}{% endif %}{% if system_message != false %}{{ '<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>' + system_message + '<|END_OF_TURN_TOKEN|>' }}{% endif %}{% for message in loop_messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% set content = message['content'] %}{% if message['role'] == 'user' %}{{ '<|START_OF_TURN_TOKEN|><|USER_TOKEN|>' + content.strip() + '<|END_OF_TURN_TOKEN|>' }}{% elif message['role'] == 'assistant' %}{{ '<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>'  + content.strip() + '<|END_OF_TURN_TOKEN|>' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>' }}{% endif %}",
     }
 
     if user_choice in templates:

src/axolotl/utils/config/models/input/v0_4_1/__init__.py

@@ -1,6 +1,7 @@
 """
 Module for pydantic models for configuration
 """
+
 # pylint: disable=too-many-lines
 
 import logging
@@ -140,6 +141,7 @@ class ChatTemplate(str, Enum):
     chatml = "chatml"  # pylint: disable=invalid-name
     inst = "inst"  # pylint: disable=invalid-name
     gemma = "gemma"  # pylint: disable=invalid-name
+    cohere = "cohere"  # pylint: disable=invalid-name
 
 
 class LoftQConfig(BaseModel):
@@ -240,17 +242,6 @@ class LoraConfig(BaseModel):
                     raise ValueError("Require cfg.load_in_4bit to be True for qlora")
         return self
 
-    @model_validator(mode="before")
-    @classmethod
-    def validate_quantized_dora(cls, data):
-        if data.get("peft_use_dora") and (
-            data.get("load_in_8bit") or data.get("load_in_4bit")
-        ):
-            raise ValueError(
-                "`peft_use_dora` is not currently compatible with quantized weights."
-            )
-        return data
-
 
 class ReLoRAConfig(BaseModel):
     """ReLoRA configuration subset"""
@@ -654,6 +645,20 @@ class AxolotlInputConfig(
 
         return data
 
+    @model_validator(mode="before")
+    @classmethod
+    def check_sample_packing_wo_flash(cls, data):
+        if (
+            data.get("sample_packing")
+            and not data.get("flash_attention")
+            and not data.get("sdp_attention")
+        ):
+            LOG.warning(
+                "sample_packing without flash_attention or sdp_attention does not handle cross-attention."
+            )
+
+        return data
+
     @model_validator(mode="before")
     @classmethod
     def check_sample_packing_w_rl(cls, data):

src/axolotl/utils/models.py

@@ -902,7 +902,12 @@ def load_lora(model, cfg, inference=False, config_only=False):
         model = get_peft_model(model, lora_config)
 
     if rank == 0:
-        model.print_trainable_parameters()
+        try:
+            model.print_trainable_parameters()
+        except AttributeError as exc:
+            LOG.warning(
+                "Exception caught during model.print_trainable_parameters(): %s", exc
+            )
     elif cfg.fsdp and cfg.adapter == "qlora":
         setup_quantized_peft_meta_for_training(model)
 

tests/e2e/patched/test_llama_s2_attention.py

@@ -7,6 +7,8 @@ import os
 import unittest
 from pathlib import Path
 
+import pytest
+
 from axolotl.cli import load_datasets
 from axolotl.common.cli import TrainerCliArgs
 from axolotl.train import train
@@ -19,6 +21,7 @@ LOG = logging.getLogger("axolotl.tests.e2e")
 os.environ["WANDB_DISABLED"] = "true"
 
 
+@pytest.mark.skip("Skipping test due to timeout.")
 class TestLlamaShiftedSparseAttention(unittest.TestCase):
     """
     Test case for Llama models using S2 Attn

tests/test_validation.py

@@ -600,6 +600,7 @@ class TestValidation(BaseValidation):
                 {
                     "sample_packing": True,
                     "pad_to_sequence_len": None,
+                    "flash_attention": True,
                 }
             )
             | minimal_cfg
@@ -901,6 +902,7 @@ class TestValidation(BaseValidation):
                 {
                     "sample_packing": True,
                     "eval_table_size": 100,
+                    "flash_attention": True,
                 }
             )
             | minimal_cfg
@@ -916,6 +918,7 @@ class TestValidation(BaseValidation):
                 {
                     "sample_packing": True,
                     "eval_sample_packing": False,
+                    "flash_attention": True,
                 }
             )
             | minimal_cfg
@@ -928,6 +931,7 @@ class TestValidation(BaseValidation):
                 {
                     "sample_packing": False,
                     "eval_table_size": 100,
+                    "flash_attention": True,
                 }
             )
             | minimal_cfg
@@ -941,6 +945,7 @@ class TestValidation(BaseValidation):
                     "sample_packing": True,
                     "eval_table_size": 100,
                     "eval_sample_packing": False,
+                    "flash_attention": True,
                 }
             )
             | minimal_cfg