Add llama2-13b-chat-v1 support to chat_models.BedrockChat (#13403)

Hi 👋 We are working with Llama2 on Bedrock, and would like to add it to
Langchain. We saw a [pull
request](https://github.com/langchain-ai/langchain/pull/13322) to add it
to the `llm.Bedrock` class, but since it concerns a chat model, we would
like to add it to `BedrockChat` as well.

- **Description:** Add support for Llama2 to `BedrockChat` in
`chat_models`
- **Issue:** the issue # it fixes (if applicable)
[#13316](https://github.com/langchain-ai/langchain/issues/13316)
  - **Dependencies:** any dependencies required for this change `None`
  - **Tag maintainer:** /
  - **Twitter handle:** `@SimonBockaert @WouterDurnez`

---------

Co-authored-by: wouter.durnez <wouter.durnez@showpad.com>
Co-authored-by: Simon Bockaert <simon.bockaert@showpad.com>

libs/langchain/langchain/chat_models/bedrock.py

@@ -5,6 +5,7 @@ from langchain.callbacks.manager import (
 )
 from langchain.chat_models.anthropic import convert_messages_to_prompt_anthropic
 from langchain.chat_models.base import BaseChatModel
+from langchain.chat_models.meta import convert_messages_to_prompt_llama
 from langchain.llms.bedrock import BedrockBase
 from langchain.pydantic_v1 import Extra
 from langchain.schema.messages import AIMessage, AIMessageChunk, BaseMessage
@@ -26,6 +27,8 @@ class ChatPromptAdapter:
     ) -> str:
         if provider == "anthropic":
             prompt = convert_messages_to_prompt_anthropic(messages=messages)
+        if provider == "meta":
+            prompt = convert_messages_to_prompt_llama(messages=messages)
         else:
             raise NotImplementedError(
                 f"Provider {provider} model does not support chat."

libs/langchain/langchain/chat_models/meta.py

@@ -0,0 +1,29 @@
+from typing import List
+
+from langchain.schema.messages import (
+    AIMessage,
+    BaseMessage,
+    ChatMessage,
+    HumanMessage,
+    SystemMessage,
+)
+
+
+def _convert_one_message_to_text_llama(message: BaseMessage) -> str:
+    if isinstance(message, ChatMessage):
+        message_text = f"\n\n{message.role.capitalize()}: {message.content}"
+    elif isinstance(message, HumanMessage):
+        message_text = f"[INST] {message.content} [/INST]"
+    elif isinstance(message, AIMessage):
+        message_text = f"{message.content}"
+    elif isinstance(message, SystemMessage):
+        message_text = f"<<SYS>> {message.content} <</SYS>>"
+    else:
+        raise ValueError(f"Got unknown type {message}")
+    return message_text
+
+
+def convert_messages_to_prompt_llama(messages: List[BaseMessage]) -> str:
+    return "\n".join(
+        [_convert_one_message_to_text_llama(message) for message in messages]
+    )

libs/langchain/langchain/llms/bedrock.py

@@ -72,6 +72,7 @@ class LLMInputOutputAdapter:
         "anthropic": "completion",
         "amazon": "outputText",
         "cohere": "text",
+        "meta": "generation",
     }
 
     @classmethod
@@ -81,7 +82,7 @@ class LLMInputOutputAdapter:
         input_body = {**model_kwargs}
         if provider == "anthropic":
             input_body["prompt"] = _human_assistant_format(prompt)
-        elif provider == "ai21" or provider == "cohere":
+        elif provider in ("ai21", "cohere", "meta"):
             input_body["prompt"] = prompt
         elif provider == "amazon":
             input_body = dict()
@@ -107,6 +108,8 @@ class LLMInputOutputAdapter:
             return response_body.get("completions")[0].get("data").get("text")
         elif provider == "cohere":
             return response_body.get("generations")[0].get("text")
+        elif provider == "meta":
+            return response_body.get("generation")
         else:
             return response_body.get("results")[0].get("outputText")
 

libs/langchain/tests/unit_tests/chat_models/test_bedrock.py

@@ -0,0 +1,30 @@
+"""Test Anthropic Chat API wrapper."""
+from typing import List
+
+import pytest
+
+from langchain.chat_models.meta import convert_messages_to_prompt_llama
+from langchain.schema import AIMessage, BaseMessage, HumanMessage, SystemMessage
+
+
+@pytest.mark.parametrize(
+    ("messages", "expected"),
+    [
+        ([HumanMessage(content="Hello")], "[INST] Hello [/INST]"),
+        (
+            [HumanMessage(content="Hello"), AIMessage(content="Answer:")],
+            "[INST] Hello [/INST]\nAnswer:",
+        ),
+        (
+            [
+                SystemMessage(content="You're an assistant"),
+                HumanMessage(content="Hello"),
+                AIMessage(content="Answer:"),
+            ],
+            "<<SYS>> You're an assistant <</SYS>>\n[INST] Hello [/INST]\nAnswer:",
+        ),
+    ],
+)
+def test_formatting(messages: List[BaseMessage], expected: str) -> None:
+    result = convert_messages_to_prompt_llama(messages)
+    assert result == expected