From 258944a53897c66442d19cde1b54c8b48a897f20 Mon Sep 17 00:00:00 2001
From: tara-servicenow <tara.bogavelli@servicenow.com>
Date: Thu, 26 Mar 2026 15:04:13 -0700
Subject: [PATCH 1/3] Update llm set up docs

---
 docs/experiment_setup.md | 16 +++++++++++-----
 1 file changed, 11 insertions(+), 5 deletions(-)

diff --git a/docs/experiment_setup.md b/docs/experiment_setup.md
index 05df3eba..8cdfd640 100644
--- a/docs/experiment_setup.md
+++ b/docs/experiment_setup.md
@@ -17,7 +17,15 @@
 
 ## LLM Configuration
 
-LLM setups in `.env` for evaluated models and judge models:
+### Reasoning effort/level for each LLM evaluated:
+- **gpt-oss-20b** - default (medium)
+- **gpt-oss-120b** - default (medium)
+- **qwen3.5-27b** - no thinking 
+- **sonnet 4.6** - low thinking 
+- **gpt-5-mini** - minimal thinking
+
+
+### LLM setups in `.env` for evaluated models and judge models:
 
 ```json
 EVA_MODEL_LIST='[
@@ -36,8 +44,7 @@ EVA_MODEL_LIST='[
     "litellm_params": {
       "model": "openai/<vllm served model name>",
       "api_key": "",
-      "api_base": "",
-      "reasoning_effort": "low"
+      "api_base": ""
     }
   },
   {
@@ -45,8 +52,7 @@ EVA_MODEL_LIST='[
     "litellm_params": {
       "model": "openai/<vllm served model name>",
       "api_key": "",
-      "api_base": "",
-      "reasoning_effort": "low"
+      "api_base": ""
     }
   },
   {

From fc31bcace73458bfd37bea4b8413969608baeaba Mon Sep 17 00:00:00 2001
From: Katrina <katrina.stankiewicz@servicenow.com>
Date: Thu, 26 Mar 2026 19:05:31 -0400
Subject: [PATCH 2/3] add support for self-hosted cohere

---
 src/eva/assistant/pipeline/services.py | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

diff --git a/src/eva/assistant/pipeline/services.py b/src/eva/assistant/pipeline/services.py
index 1fcdf76d..1582f081 100644
--- a/src/eva/assistant/pipeline/services.py
+++ b/src/eva/assistant/pipeline/services.py
@@ -134,6 +134,16 @@ def create_stt_service(
             ),
         )
 
+    elif model_lower == "cohere":
+        logger.info(f"Using Cohere STT: {params['model']}")
+        return OpenAISTTService(
+            api_key=api_key,
+            base_url=url,
+            model=params["model"],
+            language=Language.EN,
+            sample_rate=SAMPLE_RATE,
+        )
+
     elif model_lower.startswith("deepgram"):
         # Check if using Flux model
         if "flux" in model_lower:
@@ -207,7 +217,7 @@ def create_stt_service(
 
     else:
         raise ValueError(
-            f"Unknown STT model: {model}. Available: assemblyai, cartesia, deepgram, deepgram-flux, elevenlabs, nvidia, nvidia-baseten, openai"
+            f"Unknown STT model: {model}. Available: assemblyai, cartesia, cohere, deepgram, deepgram-flux, elevenlabs, nvidia, nvidia-baseten, openai"
         )
 
 

From 9f8cdce4ca846aedecb1ecf5be73108eb24fbc2f Mon Sep 17 00:00:00 2001
From: Katrina <katrina.stankiewicz@servicenow.com>
Date: Mon, 30 Mar 2026 17:47:36 -0400
Subject: [PATCH 3/3] add voxtral

---
 src/eva/assistant/pipeline/services.py | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/src/eva/assistant/pipeline/services.py b/src/eva/assistant/pipeline/services.py
index 1582f081..87a70af1 100644
--- a/src/eva/assistant/pipeline/services.py
+++ b/src/eva/assistant/pipeline/services.py
@@ -38,7 +38,7 @@
     SessionProperties,
 )
 from pipecat.services.openai.stt import OpenAISTTService
-from pipecat.services.openai.tts import VALID_VOICES, OpenAITTSService
+from pipecat.services.openai.tts import OpenAITTSService
 from pipecat.services.stt_service import STTService
 from pipecat.services.tts_service import TTSService
 from pipecat.transcriptions.language import Language
@@ -351,6 +351,18 @@ def create_tts_service(
 
         return openai_tts
 
+    elif model_lower == "voxtral":
+        logger.info(f"Using Voxtral TTS: {params['model']}")
+        voxtral_tts = OpenAITTSService(
+            api_key=api_key,
+            model=params["model"],
+            voice=params.get("voice", "neutral_female"),
+            base_url=url,
+        )
+        OpenAITTSService.run_tts = override_run_tts
+        voxtral_tts._settings.language = language_code
+        return voxtral_tts
+
     elif model_lower == "xtts":
         logger.info(f"Using XTTS TTS: {params['model']}")
         xtts_tts = OpenAITTSService(
@@ -545,7 +557,7 @@ async def override_run_tts(self, text: str, context_id: str) -> AsyncGenerator[F
         create_params = {
             "input": text,
             "model": self._settings.model,
-            "voice": VALID_VOICES[self._settings.voice],
+            "voice": self._settings.voice,
             "response_format": "pcm",
             "extra_body": {
                 "streaming_quality": "fast",