github/home-llm
1
1
Fork 0
mirror of https://github.com/acon96/home-llm.git synced 2026-01-08 21:28:05 -05:00
Code Issues Packages Projects Releases Wiki Activity

Remove Think Blocks to enhance support for chain-of-thought language models (#247)

Browse Source
This commit is contained in:
Fixt
2025-01-31 08:54:42 -05:00
committed by GitHub
parent 7c1a61c03e
commit ccd073775e
2 changed files with 13 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
custom_components/llama_conversation/const.py
View File

@@ -146,59 +146,69 @@ PROMPT_TEMPLATE_DESCRIPTIONS = {
"user": { "prefix": "<|im_start|>user\n", "suffix": "<|im_end|>" },
"assistant": { "prefix": "<|im_start|>assistant\n", "suffix": "<|im_end|>" },
"tool": { "prefix": "<|im_start|>tool", "suffix": "<|im_end|>" },
"chain_of_thought": { "prefix": "<think>", "suffix": "</think>"},
"generation_prompt": "<|im_start|>assistant"
},
PROMPT_TEMPLATE_COMMAND_R: {
"system": { "prefix": "<|START_OF_TURN_TOKEN|><|SYSTEM_TOKEN|>", "suffix": "<|END_OF_TURN_TOKEN|>" },
"user": { "prefix": "<|START_OF_TURN_TOKEN|><|USER_TOKEN|>", "suffix": "<|END_OF_TURN_TOKEN|>" },
"assistant": { "prefix": "<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>", "suffix": "<|END_OF_TURN_TOKEN|>" },
"chain_of_thought": { "prefix": "<think>", "suffix": "</think>"},
"generation_prompt": "<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>"
},
PROMPT_TEMPLATE_ALPACA: {
"system": { "prefix": "", "suffix": "\n" },
"user": { "prefix": "### Instruction:\n", "suffix": "\n" },
"assistant": { "prefix": "### Response:\n", "suffix": "\n" },
"chain_of_thought": { "prefix": "<think>", "suffix": "</think>"},
"generation_prompt": "### Response:"
},
PROMPT_TEMPLATE_VICUNA: {
"system": { "prefix": "", "suffix": "\n" },
"user": { "prefix": "USER: ", "suffix": "" },
"assistant": { "prefix": "ASSISTANT: ", "suffix": "</s>" },
"chain_of_thought": { "prefix": "<think>", "suffix": "</think>"},
"generation_prompt": "ASSISTANT:"
},
PROMPT_TEMPLATE_NONE: {
"system": { "prefix": "", "suffix": "" },
"user": { "prefix": "", "suffix": "" },
"assistant": { "prefix": "", "suffix": "" },
"chain_of_thought": { "prefix": "", "suffix": ""},
"generation_prompt": ""
},
PROMPT_TEMPLATE_MISTRAL: {
"user": { "prefix": "<s>[INST] ", "suffix": " [/INST] " },
"assistant": { "prefix": "", "suffix": "</s>" },
"chain_of_thought": { "prefix": "<think>", "suffix": "</think>"},
"generation_prompt": ""
},
PROMPT_TEMPLATE_ZEPHYR: {
"system": { "prefix": "<|system|>\n", "suffix": "<|endoftext|>" },
"user": { "prefix": "<|user|>\n", "suffix": "<|endoftext|>" },
"assistant": { "prefix": "<|assistant|>\n", "suffix": "<|endoftext|>" },
"chain_of_thought": { "prefix": "<think>", "suffix": "</think>"},
"generation_prompt": "<|assistant|>\n"
},
PROMPT_TEMPLATE_ZEPHYR2: {
"system": { "prefix": "<|system|>\n", "suffix": "</s>" },
"user": { "prefix": "<|user|>\n", "suffix": "</s>" },
"assistant": { "prefix": "<|assistant|>\n", "suffix": "</s>" },
"chain_of_thought": { "prefix": "<think>", "suffix": "</think>"},
"generation_prompt": "<|assistant|>\n"
},
PROMPT_TEMPLATE_ZEPHYR3: {
"system": { "prefix": "<|system|>\n", "suffix": "<|end|>" },
"user": { "prefix": "<|user|>\n", "suffix": "<|end|>" },
"assistant": { "prefix": "<|assistant|>\n", "suffix": "<|end|>" },
"chain_of_thought": { "prefix": "<think>", "suffix": "</think>"},
"generation_prompt": "<|assistant|>\n"
},
PROMPT_TEMPLATE_LLAMA3: {
"system": { "prefix": "<|start_header_id|>system<|end_header_id|>\n\n", "suffix": "<|eot_id|>"},
"user": { "prefix": "<|start_header_id|>user<|end_header_id|>\n\n", "suffix": "<|eot_id|>"},
"assistant": { "prefix": "<|start_header_id|>assistant<|end_header_id|>\n\n", "suffix": "<|eot_id|>"},
"chain_of_thought": { "prefix": "<think>", "suffix": "</think>"},
"generation_prompt": "<|start_header_id|>assistant<|end_header_id|>\n\n"
}
}

3
custom_components/llama_conversation/conversation.py
View File

@@ -406,6 +406,9 @@ class LocalLLMAgent(ConversationEntity, AbstractConversationAgent):
# remove end of text token if it was returned
response = response.replace(template_desc["assistant"]["suffix"], "")
# remove think blocks
response = re.sub(rf"^.*?{template_desc["chain_of_thought"]["suffix"]}", "", response, flags=re.DOTALL)
conversation.append({"role": "assistant", "message": response})
if remember_conversation:
if remember_num_interactions and len(conversation) > (remember_num_interactions * 2) + 1: