SystemSh0cker/private-gpt
1
0
Fork 0
mirror of https://github.com/zylon-ai/private-gpt.git synced 2025-12-22 07:40:12 +01:00
Code Issues Projects Releases Packages Activity

feat: prompt_style applied to all LLMs + extra LLM params. (#1835)
Some checks are pending
release-please / release-please (push) Waiting to run
tests / setup (push) Waiting to run
tests / ${{ matrix.quality-command }} (black) (push) Blocked by required conditions
tests / ${{ matrix.quality-command }} (mypy) (push) Blocked by required conditions
tests / ${{ matrix.quality-command }} (ruff) (push) Blocked by required conditions
tests / test (push) Blocked by required conditions
tests / all_checks_passed (push) Blocked by required conditions

Browse source 
* Updated prompt_style to be moved to the main LLM setting since all LLMs from llama_index can utilize this.  I also included temperature, context window size, max_tokens, max_new_tokens into the openailike to help ensure the settings are consistent from the other implementations.

* Removed prompt_style from llamacpp entirely

* Fixed settings-local.yaml to include prompt_style in the LLM settings instead of llamacpp.
This commit is contained in:
icsy7867 2024-04-30 03:53:10 -04:00 • committed by GitHub
parent c1802e7cf0
commit e21bf20c10
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 22 additions and 18 deletions
Download patch file Download diff file Expand all files Collapse all files

23
private_gpt/settings/settings.py
View file

@ -104,6 +104,17 @@ class LLMSettings(BaseModel):
0.1,
description="The temperature of the model. Increasing the temperature will make the model answer more creatively. A value of 0.1 would be more factual.",
)
prompt_style: Literal["default", "llama2", "tag", "mistral", "chatml"] = Field(
"llama2",
description=(
"The prompt style to use for the chat engine. "
"If `default` - use the default prompt style from the llama_index. It should look like `role: message`.\n"
"If `llama2` - use the llama2 prompt style from the llama_index. Based on `<s>`, `[INST]` and `<<SYS>>`.\n"
"If `tag` - use the `tag` prompt style. It should look like `<|role|>: message`. \n"
"If `mistral` - use the `mistral prompt style. It shoudl look like <s>[INST] {System Prompt} [/INST]</s>[INST] { UserInstructions } [/INST]"
"`llama2` is the historic behaviour. `default` might work better with your custom models."
),
)
class VectorstoreSettings(BaseModel):
@ -117,18 +128,6 @@ class NodeStoreSettings(BaseModel):
class LlamaCPPSettings(BaseModel):
llm_hf_repo_id: str
llm_hf_model_file: str
prompt_style: Literal["default", "llama2", "tag", "mistral", "chatml"] = Field(
"llama2",
description=(
"The prompt style to use for the chat engine. "
"If `default` - use the default prompt style from the llama_index. It should look like `role: message`.\n"
"If `llama2` - use the llama2 prompt style from the llama_index. Based on `<s>`, `[INST]` and `<<SYS>>`.\n"
"If `tag` - use the `tag` prompt style. It should look like `<|role|>: message`. \n"
"If `mistral` - use the `mistral prompt style. It shoudl look like <s>[INST] {System Prompt} [/INST]</s>[INST] { UserInstructions } [/INST]"
"`llama2` is the historic behaviour. `default` might work better with your custom models."
),
)
tfs_z: float = Field(
1.0,
description="Tail free sampling is used to reduce the impact of less probable tokens from the output. A higher value (e.g., 2.0) will reduce the impact more, while a value of 1.0 disables this setting.",