Merge pull request #266 from ServiceNow/new_experiments

Adding new llms

Author: amanjaiswal73892

.vscode/settings.json

@@ -7,8 +7,8 @@
             "source.fixAll": "always",
         },
     },
-    "python.analysis.languageServerMode": "full",
-    "python.analysis.typeCheckingMode": "standard",
+    // "python.analysis.languageServerMode": "full",
+    // "python.analysis.typeCheckingMode": "standard",
     "python.testing.pytestArgs": [
         "tests"
     ],

src/agentlab/llm/chat_api.py

@@ -9,7 +9,7 @@
 import anthropic
 import openai
 from huggingface_hub import InferenceClient
-from openai import AzureOpenAI, OpenAI
+from openai import NOT_GIVEN, AzureOpenAI, OpenAI
 
 import agentlab.llm.tracking as tracking
 from agentlab.llm.base_api import AbstractChatModel, BaseModelArgs
@@ -348,6 +348,8 @@ def __init__(
         min_retry_wait_time=60,
         log_probs=False,
     ):
+        if max_tokens is None:
+            max_tokens = NOT_GIVEN
         super().__init__(
             model_name=model_name,
             api_key=api_key,
@@ -482,13 +484,11 @@ def __init__(
         temperature=0.5,
         max_tokens=100,
         max_retry=4,
-        log_probs=False,
     ):
         self.model_name = model_name
         self.temperature = temperature
         self.max_tokens = max_tokens
         self.max_retry = max_retry
-        self.log_probs = log_probs
 
         api_key = api_key or os.getenv("ANTHROPIC_API_KEY")
         self.client = anthropic.Anthropic(api_key=api_key)
@@ -544,5 +544,4 @@ def make_model(self):
             model_name=self.model_name,
             temperature=self.temperature,
             max_tokens=self.max_new_tokens,
-            log_probs=self.log_probs,
         )

src/agentlab/llm/llm_configs.py

@@ -1,9 +1,11 @@
 from agentlab.llm.chat_api import (
+    AnthropicModelArgs,
     AzureModelArgs,
     OpenAIModelArgs,
     OpenRouterModelArgs,
     SelfHostedModelArgs,
 )
+from openai import NOT_GIVEN
 
 default_oss_llms_args = {
     "n_retry_server": 4,
@@ -38,6 +40,14 @@
         max_new_tokens=100_000,
         vision_support=False,
     ),
+    "openai/o3-2025-04-16": OpenAIModelArgs(
+        model_name="o3-2025-04-16",
+        max_total_tokens=200_000,
+        max_input_tokens=200_000,
+        max_new_tokens=None,
+        temperature=1,
+        vision_support=True,
+    ),
     "openai/gpt-4o-mini-2024-07-18": OpenAIModelArgs(
         model_name="gpt-4o-mini-2024-07-18",
         max_total_tokens=128_000,
@@ -115,6 +125,17 @@
         max_new_tokens=16_384,
         vision_support=True,
     ),
+    # ---------------- Anthropic ----------------#
+    "anthropic/claude-3-7-sonnet-20250219": AnthropicModelArgs(
+        model_name="claude-3-7-sonnet-20250219",
+        max_new_tokens=16_384,
+        temperature=1e-1,
+    ),
+    "anthropic/claude-sonnet-4-20250514": AnthropicModelArgs(
+        model_name="claude-sonnet-4-20250514",
+        max_new_tokens=16_384,
+        temperature=1e-1,
+    ),
     # ---------------- OSS LLMs ----------------#
     "meta-llama/Meta-Llama-3-70B-Instruct": SelfHostedModelArgs(
         model_name="meta-llama/Meta-Llama-3-70B-Instruct",