feat: make llama3.1 as default (#2022)

* feat: change ollama default model to llama3.1

* chore: bump versions

* feat: Change default model in local mode to llama3.1

* chore: make sure last poetry version is used

* fix: mypy

* fix: do not add BOS (with last llamacpp-python version)

.github/workflows/actions/install_dependencies/action.yml

@@ -8,7 +8,7 @@ inputs:
   poetry_version:
     required: true
     type: string
-    default: "1.5.1"
+    default: "1.8.3"
 
 runs:
   using: composite

Dockerfile.external

@@ -2,7 +2,8 @@ FROM python:3.11.6-slim-bookworm as base
 
 # Install poetry
 RUN pip install pipx
-RUN pipx install poetry
+RUN python3 -m pipx ensurepath
+RUN pipx install poetry==1.8.3
 ENV PATH="/root/.local/bin:$PATH"
 ENV PATH=".venv/bin/:$PATH"
 

Dockerfile.local

@@ -4,7 +4,8 @@ FROM python:3.11.6-slim-bookworm as base
 
 # Install poetry
 RUN pip install pipx
-RUN pipx install poetry
+RUN python3 -m pipx ensurepath
+RUN pipx install poetry==1.8.3
 ENV PATH="/root/.local/bin:$PATH"
 ENV PATH=".venv/bin/:$PATH"
 

fern/docs/pages/installation/installation.mdx

@@ -28,6 +28,11 @@ pyenv local 3.11
 Install [Poetry](https://python-poetry.org/docs/#installing-with-the-official-installer) for dependency management:
 Follow the instructions on the official Poetry website to install it.
 
+<Callout intent="warning">
+A bug exists in Poetry versions 1.7.0 and earlier. We strongly recommend upgrading to a tested version.
+To upgrade Poetry to latest tested version, run `poetry self update 1.8.3` after installing it.
+</Callout>
+
 ### 4. Optional: Install `make`
 To run various scripts, you need to install `make`. Follow the instructions for your operating system:
 #### macOS
@@ -135,14 +140,14 @@ Now, start Ollama service (it will start a local inference server, serving both
 ollama serve
 ```
 
-Install the models to be used, the default settings-ollama.yaml is configured to user mistral 7b LLM (~4GB) and nomic-embed-text Embeddings (~275MB)
+Install the models to be used, the default settings-ollama.yaml is configured to user llama3.1 8b LLM (~4GB) and nomic-embed-text Embeddings (~275MB)
 
 By default, PGPT will automatically pull models as needed. This behavior can be changed by modifying the `ollama.autopull_models` property.
 
 In any case, if you want to manually pull models, run the following commands:
 
 ```bash
-ollama pull mistral
+ollama pull llama3.1
 ollama pull nomic-embed-text
 ```
 

fern/docs/pages/installation/troubleshooting.mdx

@@ -24,7 +24,7 @@ PrivateGPT uses the `AutoTokenizer` library to tokenize input text accurately. I
    In your `settings.yaml` file, specify the model you want to use:
    ```yaml
    llm:
-     tokenizer: mistralai/Mistral-7B-Instruct-v0.2
+     tokenizer: meta-llama/Meta-Llama-3.1-8B-Instruct
    ```
 2. **Set Access Token for Gated Models:**
    If you are using a gated model, ensure the `access_token` is set as mentioned in the previous section.

private_gpt/components/llm/prompt_helper.py

@@ -169,7 +169,7 @@ class Llama3PromptStyle(AbstractPromptStyle):
     """
 
     def _messages_to_prompt(self, messages: Sequence[ChatMessage]) -> str:
-        prompt = self.BOS
+        prompt = ""
         has_system_message = False
 
         for i, message in enumerate(messages):
@@ -189,8 +189,7 @@ def _messages_to_prompt(self, messages: Sequence[ChatMessage]) -> str:
         # Add default system prompt if no system message was provided
         if not has_system_message:
             prompt = (
-                f"{self.BOS}{self.B_SYS}\n\n{self.DEFAULT_SYSTEM_PROMPT}{self.E_SYS}"
-                + prompt[len(self.BOS) :]
+                f"{self.B_SYS}\n\n{self.DEFAULT_SYSTEM_PROMPT}{self.E_SYS}" + prompt
             )
 
         # TODO: Implement tool handling logic
@@ -199,7 +198,7 @@ def _messages_to_prompt(self, messages: Sequence[ChatMessage]) -> str:
 
     def _completion_to_prompt(self, completion: str) -> str:
         return (
-            f"{self.BOS}{self.B_SYS}\n\n{self.DEFAULT_SYSTEM_PROMPT}{self.E_SYS}"
+            f"{self.B_SYS}\n\n{self.DEFAULT_SYSTEM_PROMPT}{self.E_SYS}"
             f"{self.B_INST}user{self.E_INST}\n\n{completion.strip()}{self.EOT}"
             f"{self.ASSISTANT_INST}\n\n"
         )

private_gpt/launcher.py

@@ -37,7 +37,8 @@ async def bind_injector_to_request(request: Request) -> None:
 
     # Add LlamaIndex simple observability
     global_handler = create_global_handler("simple")
-    LlamaIndexSettings.callback_manager = CallbackManager([global_handler])
+    if global_handler is not None:
+        LlamaIndexSettings.callback_manager = CallbackManager([global_handler])
 
     settings = root_injector.get(Settings)
     if settings.server.cors.enabled:

settings-docker.yaml

@@ -9,8 +9,8 @@ embedding:
   mode: ${PGPT_EMBED_MODE:mock}
 
 llamacpp:
-  llm_hf_repo_id: ${PGPT_HF_REPO_ID:TheBloke/Mistral-7B-Instruct-v0.1-GGUF}
-  llm_hf_model_file: ${PGPT_HF_MODEL_FILE:mistral-7b-instruct-v0.1.Q4_K_M.gguf}
+  llm_hf_repo_id: ${PGPT_HF_REPO_ID:lmstudio-community/Meta-Llama-3.1-8B-Instruct-GGUF}
+  llm_hf_model_file: ${PGPT_HF_MODEL_FILE:Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf}
 
 huggingface:
   embedding_hf_model_name: ${PGPT_EMBEDDING_HF_MODEL_NAME:BAAI/bge-small-en-v1.5}
@@ -20,7 +20,7 @@ sagemaker:
   embedding_endpoint_name: ${PGPT_SAGEMAKER_EMBEDDING_ENDPOINT_NAME:}
 
 ollama:
-  llm_model: ${PGPT_OLLAMA_LLM_MODEL:mistral}
+  llm_model: ${PGPT_OLLAMA_LLM_MODEL:llama3.1}
   embedding_model: ${PGPT_OLLAMA_EMBEDDING_MODEL:nomic-embed-text}
   api_base: ${PGPT_OLLAMA_API_BASE:http://ollama:11434}
   embedding_api_base: ${PGPT_OLLAMA_EMBEDDING_API_BASE:http://ollama:11434}

settings-local.yaml

@@ -7,12 +7,12 @@ llm:
   # Should be matching the selected model
   max_new_tokens: 512
   context_window: 3900
-  tokenizer: mistralai/Mistral-7B-Instruct-v0.2
-  prompt_style: "mistral"
+  tokenizer: meta-llama/Meta-Llama-3.1-8B-Instruct
+  prompt_style: "llama3"
 
 llamacpp:
-  llm_hf_repo_id: TheBloke/Mistral-7B-Instruct-v0.2-GGUF
-  llm_hf_model_file: mistral-7b-instruct-v0.2.Q4_K_M.gguf
+  llm_hf_repo_id: lmstudio-community/Meta-Llama-3.1-8B-Instruct-GGUF
+  llm_hf_model_file: Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf
 
 embedding:
   mode: huggingface

settings-ollama-pg.yaml

@@ -14,7 +14,7 @@ embedding:
   embed_dim: 768
 
 ollama:
-  llm_model: mistral
+  llm_model: llama3.1
   embedding_model: nomic-embed-text
   api_base: http://localhost:11434
 

settings-ollama.yaml

@@ -11,7 +11,7 @@ embedding:
   mode: ollama
 
 ollama:
-  llm_model: mistral
+  llm_model: llama3.1
   embedding_model: nomic-embed-text
   api_base: http://localhost:11434
   embedding_api_base: http://localhost:11434  # change if your embedding model runs on another ollama

settings-vllm.yaml

@@ -4,7 +4,7 @@ server:
 llm:
   mode: openailike
   max_new_tokens: 512
-  tokenizer: mistralai/Mistral-7B-Instruct-v0.2
+  tokenizer: meta-llama/Meta-Llama-3.1-8B-Instruct
   temperature: 0.1
 
 embedding:

settings.yaml

@@ -39,12 +39,12 @@ ui:
 
 llm:
   mode: llamacpp
-  prompt_style: "mistral"
+  prompt_style: "llama3"
   # Should be matching the selected model
   max_new_tokens: 512
   context_window: 3900
   # Select your tokenizer. Llama-index tokenizer is the default.
-  # tokenizer: mistralai/Mistral-7B-Instruct-v0.2
+  # tokenizer: meta-llama/Meta-Llama-3.1-8B-Instruct
   temperature: 0.1      # The temperature of the model. Increasing the temperature will make the model answer more creatively. A value of 0.1 would be more factual. (Default: 0.1)
 
 rag:
@@ -65,8 +65,8 @@ clickhouse:
     database: embeddings
 
 llamacpp:
-  llm_hf_repo_id: TheBloke/Mistral-7B-Instruct-v0.2-GGUF
-  llm_hf_model_file: mistral-7b-instruct-v0.2.Q4_K_M.gguf
+  llm_hf_repo_id: lmstudio-community/Meta-Llama-3.1-8B-Instruct-GGUF
+  llm_hf_model_file: Meta-Llama-3.1-8B-Instruct-Q4_K_M.gguf
   tfs_z: 1.0            # 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
   top_k: 40             # Reduces the probability of generating nonsense. A higher value (e.g. 100) will give more diverse answers, while a lower value (e.g. 10) will be more conservative. (Default: 40)
   top_p: 1.0            # Works together with top-k. A higher value (e.g., 0.95) will lead to more diverse text, while a lower value (e.g., 0.5) will generate more focused and conservative text. (Default: 0.9)
@@ -114,7 +114,7 @@ openai:
   embedding_api_key: ${OPENAI_API_KEY:}
 
 ollama:
-  llm_model: llama2
+  llm_model: llama3.1
   embedding_model: nomic-embed-text
   api_base: http://localhost:11434
   embedding_api_base: http://localhost:11434  # change if your embedding model runs on another ollama

tests/test_prompt_helper.py

@@ -150,7 +150,7 @@ def test_llama3_prompt_style_format():
     ]
 
     expected_prompt = (
-        "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\n"
+        "<|start_header_id|>system<|end_header_id|>\n\n"
         "You are a helpful assistant<|eot_id|>"
         "<|start_header_id|>user<|end_header_id|>\n\n"
         "Hello, how are you doing?<|eot_id|>"
@@ -166,7 +166,7 @@ def test_llama3_prompt_style_with_default_system():
         ChatMessage(content="Hello!", role=MessageRole.USER),
     ]
     expected = (
-        "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\n"
+        "<|start_header_id|>system<|end_header_id|>\n\n"
         f"{prompt_style.DEFAULT_SYSTEM_PROMPT}<|eot_id|>"
         "<|start_header_id|>user<|end_header_id|>\n\nHello!<|eot_id|>"
         "<|start_header_id|>assistant<|end_header_id|>\n\n"
@@ -185,7 +185,7 @@ def test_llama3_prompt_style_with_assistant_response():
     ]
 
     expected_prompt = (
-        "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\n"
+        "<|start_header_id|>system<|end_header_id|>\n\n"
         "You are a helpful assistant<|eot_id|>"
         "<|start_header_id|>user<|end_header_id|>\n\n"
         "What is the capital of France?<|eot_id|>"