tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Multimodal Vision Llama - rudimentary support (#1940)

Browse Source 
---------

Co-authored-by: Sunny <sunny@Sunnys-MacBook-Air.local>
Co-authored-by: sunny <sunnyliu19981005@gmail.com>
This commit is contained in:
Wing Lian
2024-10-02 21:02:48 -04:00
committed by GitHub
parent 844331005c
commit e1915f5625
24 changed files with 799 additions and 119 deletions
Download Patch File Download Diff File Expand all files Collapse all files

21
tests/prompt_strategies/test_chat_templates.py
View File

@@ -73,7 +73,7 @@ class TestAssistantChatTemplateLlama3:
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(
llama3_tokenizer,
chat_templates("llama3"),
chat_template=chat_templates("llama3"),
message_field_role="role",
message_field_content="content",
roles={
@@ -113,7 +113,7 @@ class TestAssistantChatTemplateLlama3:
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(
phi35_tokenizer,
chat_templates("phi_35"),
chat_template=chat_templates("phi_35"),
message_field_role="role",
message_field_content="content",
roles={
@@ -171,7 +171,7 @@ class TestAssistantChatTemplateLlama3:
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(
llama3_tokenizer,
chat_templates("llama3"),
chat_template=chat_templates("llama3"),
message_field_role="role",
message_field_content="content",
message_field_training="training",
@@ -227,8 +227,11 @@ class TestSharegptChatTemplateLlama3:
def test_llama3_assistant(self, llama3_tokenizer, sharegpt_dataset):
LOG.info("Testing ShareGPT style datasets with llama-3 assistant prompts")
# pylint: disable=duplicate-code
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(llama3_tokenizer, chat_templates("llama3")),
ChatTemplatePrompter(
llama3_tokenizer, chat_template=chat_templates("llama3")
),
tokenizer=llama3_tokenizer,
train_on_inputs=False,
train_on_eos="none",
@@ -277,8 +280,11 @@ class TestSharegptChatTemplateLlama3:
def test_llama3_human(self, llama3_tokenizer, sharegpt_dataset):
LOG.info("Testing ShareGPT style datasets with llama-3 human prompts")
# pylint: disable=duplicate-code
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(llama3_tokenizer, chat_templates("llama3")),
ChatTemplatePrompter(
llama3_tokenizer, chat_template=chat_templates("llama3")
),
tokenizer=llama3_tokenizer,
train_on_inputs=False,
train_on_eos="none",
@@ -327,8 +333,11 @@ class TestSharegptChatTemplateLlama3:
def test_llama3_system_human(self, llama3_tokenizer, basic_dataset):
LOG.info("Testing ShareGPT style datasets with llama-3 system/human prompts")
# pylint: disable=duplicate-code
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(llama3_tokenizer, chat_templates("llama3")),
ChatTemplatePrompter(
llama3_tokenizer, chat_template=chat_templates("llama3")
),
tokenizer=llama3_tokenizer,
train_on_inputs=False,
train_on_eos="none",

46
tests/prompt_strategies/test_chat_templates_advanced.py
View File

@@ -34,7 +34,9 @@ class TestChatTemplateConfigurations:
def test_train_on_inputs_true(self, llama3_tokenizer, basic_dataset):
LOG.info("Testing with train_on_inputs=True")
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(llama3_tokenizer, chat_templates("llama3")),
ChatTemplatePrompter(
llama3_tokenizer, chat_template=chat_templates("llama3")
),
tokenizer=llama3_tokenizer,
train_on_inputs=True,
sequence_len=512,
@@ -77,7 +79,9 @@ class TestChatTemplateConfigurations:
def test_train_on_inputs_false(self, llama3_tokenizer, basic_dataset):
LOG.info("Testing with train_on_inputs=False")
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(llama3_tokenizer, chat_templates("llama3")),
ChatTemplatePrompter(
llama3_tokenizer, chat_template=chat_templates("llama3")
),
tokenizer=llama3_tokenizer,
train_on_inputs=False,
sequence_len=512,
@@ -118,7 +122,9 @@ class TestChatTemplateConfigurations:
def test_roles_to_train_assistant_only(self, llama3_tokenizer, basic_dataset):
LOG.info("Testing roles_to_train with assistant only")
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(llama3_tokenizer, chat_templates("llama3")),
ChatTemplatePrompter(
llama3_tokenizer, chat_template=chat_templates("llama3")
),
tokenizer=llama3_tokenizer,
train_on_inputs=False,
sequence_len=512,
@@ -144,7 +150,9 @@ class TestChatTemplateConfigurations:
def test_roles_to_train_all(self, llama3_tokenizer, basic_dataset):
LOG.info("Testing roles_to_train with all roles")
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(llama3_tokenizer, chat_templates("llama3")),
ChatTemplatePrompter(
llama3_tokenizer, chat_template=chat_templates("llama3")
),
tokenizer=llama3_tokenizer,
train_on_inputs=True,
sequence_len=512,
@@ -175,7 +183,9 @@ class TestChatTemplateConfigurations:
def test_empty_roles_to_train(self, llama3_tokenizer, basic_dataset):
LOG.info("Testing with empty roles_to_train")
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(llama3_tokenizer, chat_templates("llama3")),
ChatTemplatePrompter(
llama3_tokenizer, chat_template=chat_templates("llama3")
),
tokenizer=llama3_tokenizer,
train_on_inputs=False,
sequence_len=512,
@@ -194,7 +204,9 @@ class TestChatTemplateConfigurations:
def test_train_on_eos_all(self, llama3_tokenizer, basic_dataset):
LOG.info("Testing with train_on_eos='all'")
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(llama3_tokenizer, chat_templates("llama3")),
ChatTemplatePrompter(
llama3_tokenizer, chat_template=chat_templates("llama3")
),
tokenizer=llama3_tokenizer,
train_on_inputs=False,
sequence_len=512,
@@ -219,7 +231,9 @@ class TestChatTemplateConfigurations:
def test_train_on_eos_turn(self, llama3_tokenizer, basic_dataset):
LOG.info("Testing with train_on_eos='turn'")
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(llama3_tokenizer, chat_templates("llama3")),
ChatTemplatePrompter(
llama3_tokenizer, chat_template=chat_templates("llama3")
),
tokenizer=llama3_tokenizer,
train_on_inputs=False,
sequence_len=512,
@@ -267,7 +281,9 @@ class TestChatTemplateConfigurations:
def test_train_on_eos_last(self, llama3_tokenizer, basic_dataset):
LOG.info("Testing with train_on_eos='last'")
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(llama3_tokenizer, chat_templates("llama3")),
ChatTemplatePrompter(
llama3_tokenizer, chat_template=chat_templates("llama3")
),
tokenizer=llama3_tokenizer,
train_on_inputs=False,
sequence_len=512,
@@ -298,7 +314,9 @@ class TestChatTemplateConfigurations:
def test_train_on_eos_none(self, llama3_tokenizer, basic_dataset):
LOG.info("Testing with train_on_eos='none'")
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(llama3_tokenizer, chat_templates("llama3")),
ChatTemplatePrompter(
llama3_tokenizer, chat_template=chat_templates("llama3")
),
tokenizer=llama3_tokenizer,
train_on_inputs=False,
sequence_len=512,
@@ -324,7 +342,9 @@ class TestChatTemplateConfigurations:
LOG.info("Testing with drop_system_message=True")
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(
llama3_tokenizer, chat_templates("llama3"), drop_system_message=True
llama3_tokenizer,
chat_template=chat_templates("llama3"),
drop_system_message=True,
),
tokenizer=llama3_tokenizer,
train_on_inputs=False,
@@ -350,7 +370,9 @@ class TestChatTemplateConfigurations:
}
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(
llama3_tokenizer, chat_templates("llama3"), roles=custom_roles
llama3_tokenizer,
chat_template=chat_templates("llama3"),
roles=custom_roles,
),
tokenizer=llama3_tokenizer,
train_on_inputs=False,
@@ -402,7 +424,7 @@ class TestChatTemplateConfigurations:
strategy = ChatTemplateStrategy(
ChatTemplatePrompter(
llama3_tokenizer,
chat_templates("llama3"),
chat_template=chat_templates("llama3"),
message_field_training="train",
message_field_training_detail="train_detail",
),