Openai api completion params [seed, temperature, max_tokens, system_fingerprint]

README.md

@@ -200,6 +200,10 @@ streamlit run torchchat.py -- browser llama3.1
 <details>
 <summary>This mode gives a REST API that matches the OpenAI API spec for interacting with a model</summary>
 
+The server follows the [OpenAI API specification](https://platform.openai.com/docs/api-reference/chat) for chat completions.
+Since this feature is under active development, it's possible not every parameter is consumed. See api/api.py for details on
+which request parameters are implemented. If you encounter any issues, please comment on the [tracking Github issue](https://github.com/pytorch/torchchat/issues/973).
+
 To test out the REST API, **you'll need 2 terminals**: one to host the server, and one to send the request.
 
 In one terminal, start the server
@@ -213,8 +217,7 @@ python3 torchchat.py server llama3.1
 
 In another terminal, query the server using `curl`. Depending on the model configuration, this query might take a few minutes to respond.
 
-Setting `stream` to "true" in the request emits a response in chunks. Currently, this response
-is plaintext and will not be formatted to the OpenAI API specification. If `stream` is unset or not "true", then the client will await the full response from the server.
+Setting `stream` to "true" in the request emits a response in chunks. If `stream` is unset or not "true", then the client will await the full response from the server.
 
 
 **Example Input + Output**
@@ -227,6 +230,7 @@ curl http://127.0.0.1:5000/v1/chat \
   -d '{
     "model": "llama3.1",
     "stream": "true",
+    "max_tokens": 200,
     "messages": [
       {
         "role": "system",

api/api.py

@@ -19,6 +19,8 @@
 See https://platform.openai.com/docs/api-reference/chat for the full specification and details.
 """
 
+OPENAI_API_DEFAULT_MAX_TOKENS = 16
+
 # Message classes and associated objects - see the types of Messages under "Create Chat Completion >>> Request body >>> messages"
 
 
@@ -105,20 +107,20 @@ class CompletionRequest:
     logit_bias: Optional[Dict[str, float]] = None  # unimplemented
     logprobs: Optional[bool] = None  # unimplemented
     top_logprobs: Optional[int] = None  # unimplemented
-    max_tokens: Optional[int] = None  # unimplemented
+    max_tokens: Optional[int] = None
     n: int = 1
     presence_penalty: float = 0  # unimplemented
     response_format: Optional[ResponseFormat] = None  # unimplemented
-    seed: Optional[int] = None  # unimplemented
+    seed: Optional[int] = None
     service_tier: Optional[str] = None  # unimplemented
     stop: Optional[List[str]] = None  # unimplemented
     stream: bool = False
     stream_options: Optional[StreamOptions] = None  # unimplemented
-    temperature: Optional[float] = 1.0  # unimplemented
+    temperature: Optional[float] = 1.0
     top_p: Optional[float] = 1.0  # unimplemented
-    tools: Optional[List[Any]] = None  # unimplemented
-    tool_choice: Optional[Union[str, Any]] = None  # unimplemented
-    parallel_tool_calls: Optional[bool] = None  # unimplemented
+    tools: Optional[List[Any]] = None  # unimplemented - Assistant features
+    tool_choice: Optional[Union[str, Any]] = None  # unimplemented - Assistant features
+    parallel_tool_calls: Optional[bool] = None  # unimplemented - Assistant features
     user: Optional[str] = None  # unimplemented
 
 
@@ -229,9 +231,8 @@ def __init__(self, *args, **kwargs):
             else self.model.config.max_seq_length
         )
         # The System fingerprint is a unique identifier for the model and its configuration.
-        # Currently, this is not implemented in a
         self.system_fingerprint = (
-            self.builder_args.device + type(self.builder_args.precision).__name__
+            f"{self.builder_args.device}_{self.builder_args.precision}"
         )
 
     def chunked_completion(self, completion_request: CompletionRequest):
@@ -270,7 +271,13 @@ def chunked_completion(self, completion_request: CompletionRequest):
         )
         generator_args = GeneratorArgs(
             completion_request.messages[-1].get("content"),
+            max_new_tokens=(
+                int(completion_request.max_tokens)
+                if completion_request.max_tokens
+                else OPENAI_API_DEFAULT_MAX_TOKENS
+            ),
             encoded_prompt=encoded,
+            temperature=float(completion_request.temperature),
             chat_mode=False,
         )
 
@@ -295,6 +302,7 @@ def callback(x, *, done_generating=False):
             sequential_prefill=generator_args.sequential_prefill,
             start_pos=self.start_pos,
             max_seq_length=self.max_seq_length,
+            seed=int(completion_request.seed),
         ):
             if y is None:
                 continue

generate.py

@@ -71,7 +71,7 @@ class GeneratorArgs:
     num_samples: int = 1
     max_new_tokens: int = 200
     top_k: int = 200
-    temperature: int = 0  # deterministic argmax
+    temperature: float = 0.0  # deterministic argmax if 0.0
     compile: bool = False
     compile_prefill: bool = False
     speculate_k: int = 5
@@ -105,9 +105,7 @@ def validate_build(
     def from_args(cls, args):
         dso_path = getattr(args, "dso_path", None)
         pte_path = getattr(args, "pte_path", None)
-        sequential_prefill = (
-            args.sequential_prefill or bool(dso_path) or bool(pte_path)
-        )
+        sequential_prefill = args.sequential_prefill or bool(dso_path) or bool(pte_path)
 
         return cls(
             prompt=getattr(args, "prompt", ""),

server.py

@@ -9,6 +9,8 @@
 from dataclasses import asdict
 from typing import Dict, List, Union
 
+import torch
+
 from api.api import CompletionRequest, OpenAiApiGenerator
 from api.models import get_model_info_list, retrieve_model_info
 
@@ -50,6 +52,8 @@ def chat_endpoint():
         """
 
         print(" === Completion Request ===")
+        if seed := request.args.get("seed"):
+            torch.manual_seed(int(seed))
 
         # Parse the request in to a CompletionRequest object
         data = request.get_json()