fix: fireworks provider chat completion failing

[slekkala1]

What does this PR do?

Tried to fix earlier, however since fireworks api returns usage in response, this propagation can fix the end to end response to have the usage
Context in #3392
Closes #3391

Test Plan

Build-run server then test with curl

(llama-stack) (base) swapna942@swapna942-mac llama-stack % curl -X POST http://localhost:8321/v1/openai/v1/chat/completions \
      -H "Content-Type: application/json" \
      -H "X-LlamaStack-Provider-Data: {\"fireworks_api_key\": \"$FIREWORKS_API_KEY\"}" \
      -d '{
        "model": "fireworks/accounts/fireworks/models/llama-v3p1-8b-instruct",
        "messages": [{"role": "user", "content": "Hello!"}]
      }' | jq
{
  "metrics": [
    {
      "metric": "prompt_tokens",
      "value": 35,
      "unit": "tokens"
    },
    {
      "metric": "completion_tokens",
      "value": 10,
      "unit": "tokens"
    },
    {
      "metric": "total_tokens",
      "value": 45,
      "unit": "tokens"
    }
  ],
  "id": "chatcmpl-0646bc72-cf4a-4afd-8fa0-030275b452f6",
  "choices": [
    {
      "message": {
        "role": "assistant",
        "content": "Hello! How can I assist you today?",
        "name": null,
        "tool_calls": null
      },
      "finish_reason": "stop",
      "index": 0,
      "logprobs": null
    }
  ],
  "object": "chat.completion",
  "created": 1757632496,
  "model": "fireworks/accounts/fireworks/models/llama-v3p1-8b-instruct",
  "usage": {
    "prompt_tokens": 35,
    "completion_tokens": 10,
    "total_tokens": 45
  }
} 

[mattf]

this needs to also revert https://github.com/llamastack/llama-stack/pull/3392/files

[mattf]

@slekkala1 we're moving all the adapter we can to the OpenAIMixin, will you convert the fireworks adapter to use OpenAIMixin and fix usage for everyone?

[slekkala1]

this needs to also revert https://github.com/llamastack/llama-stack/pull/3392/files

We revert this already @mattf #3402

[slekkala1]

@slekkala1 we're moving all the adapter we can to the OpenAIMixin, will you convert the fireworks adapter to use OpenAIMixin and fix usage for everyone?

@mattf Yes I can do that, let me look into that instead then. Please link any related task you have

I see another similar bug got reported #3420 for the response.usage

[mattf]

@slekkala1 we're moving all the adapter we can to the OpenAIMixin, will you convert the fireworks adapter to use OpenAIMixin and fix usage for everyone?

@mattf Yes I can do that, let me look into that instead then. Please link any related task you have

I see another similar bug got reported #3420 for the response.usage

#3387 has a list of other provider updates.

i expect Fireworks can end up looking like https://github.com/llamastack/llama-stack/blob/main/llama_stack/providers/remote/inference/gemini/gemini.py

however, don't bother updating the embedding/completion/chat_completion methods, they're on their way out #2365

[mattf]

this may help for testing -

diff --git a/tests/integration/suites.py b/tests/integration/suites.py
index bacd7ef5..4e653605 100644
--- a/tests/integration/suites.py
+++ b/tests/integration/suites.py
@@ -90,6 +90,15 @@ SETUP_DEFINITIONS: dict[str, Setup] = {
             "embedding_model": "sentence-transformers/all-MiniLM-L6-v2",
         },
     ),
+    "fireworks": Setup(
+        name="fireworks",
+        description="Fireworks provider with a text model",
+        defaults={
+            "text_model": "accounts/fireworks/models/llama-v3p2-3b-instruct",
+            "vision_model": "accounts/fireworks/models/llama-v3p2-11b-vision-instruct",
+            "embedding_model": "accounts/fireworks/models/qwen3-embedding-8b",
+        },
+    ),
 }

[slekkala1]

this may help for testing -

diff --git a/tests/integration/suites.py b/tests/integration/suites.py
index bacd7ef5..4e653605 100644
--- a/tests/integration/suites.py
+++ b/tests/integration/suites.py
@@ -90,6 +90,15 @@ SETUP_DEFINITIONS: dict[str, Setup] = {
             "embedding_model": "sentence-transformers/all-MiniLM-L6-v2",
         },
     ),
+    "fireworks": Setup(
+        name="fireworks",
+        description="Fireworks provider with a text model",
+        defaults={
+            "text_model": "accounts/fireworks/models/llama-v3p2-3b-instruct",
+            "vision_model": "accounts/fireworks/models/llama-v3p2-11b-vision-instruct",
+            "embedding_model": "accounts/fireworks/models/qwen3-embedding-8b",
+        },
+    ),
 }

Thanks for this, yes looking into some test failures for fireworks. Running into rate limit execeeded.

Thinking of having a separate diff once I fix some of these tests. Will close this diff soon.

[slekkala1]

@mattf Closing this diff, Opened new draft at #3480 with test and OpenAIMixin refactoring.
However, since firework provider uses OpenAIChatCompletionToLlamaStackMixin for llama_models, we continue fix this bug with response.usage in openai_compat.py