Add QwQ model on Together AI (#3307)

stanford-crfm · Feb 4, 2025 · f62ea62 · f62ea62
1 parent 5a50569
commit f62ea62
Show file tree

Hide file tree

Showing 3 changed files with 24 additions and 2 deletions.
diff --git a/src/helm/config/model_deployments.yaml b/src/helm/config/model_deployments.yaml
@@ -2824,6 +2824,13 @@ model_deployments:
     client_spec:
       class_name: "helm.clients.together_client.TogetherChatClient"
 
+  - name: together/qwq-32b-preview
+    model_name: qwen/qwq-32b-preview
+    tokenizer_name: qwen/qwq-32b-preview
+    max_sequence_length: 32768
+    client_spec:
+      class_name: "helm.clients.together_client.TogetherChatClient"
+
   - name: huggingface/qwen-vl
     model_name: qwen/qwen-vl
     tokenizer_name: qwen/qwen-vl

diff --git a/src/helm/config/model_metadata.yaml b/src/helm/config/model_metadata.yaml
@@ -2909,6 +2909,15 @@ models:
     release_date: 2024-09-19
     tags: [TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG, INSTRUCTION_FOLLOWING_MODEL_TAG]
 
+  - name: qwen/qwq-32b-preview
+    display_name: QwQ (32B Preview)
+    description: QwQ-32B-Preview is an experimental research model developed by the Qwen Team, focused on advancing AI reasoning capabilities. ([blog post](https://qwenlm.github.io/blog/qwq-32b-preview/)).
+    creator_organization_name: Alibaba Cloud
+    access: open
+    num_parameters: 32800000000
+    release_date: 2024-11-28
+    tags: [TEXT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG, INSTRUCTION_FOLLOWING_MODEL_TAG]
+
   - name: qwen/qwen-vl
     display_name: Qwen-VL
     description: Visual multimodal version of the Qwen large language model series ([paper](https://arxiv.org/abs/2308.12966)).

diff --git a/src/helm/config/tokenizer_configs.yaml b/src/helm/config/tokenizer_configs.yaml
@@ -578,15 +578,21 @@ tokenizer_configs:
       args:
         pretrained_model_name_or_path: Qwen/Qwen2-72B-Instruct
     end_of_text_token: "<|im_end|>"
-    prefix_token: "<|im_start|>'"
+    prefix_token: "<|im_start|>"
 
   - name: qwen/qwen2.5-7b-instruct
     tokenizer_spec:
       class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
       args:
         pretrained_model_name_or_path: Qwen/Qwen2.5-7B-Instruct
     end_of_text_token: "<|im_end|>"
-    prefix_token: "<|im_start|>'"
+    prefix_token: "<|im_start|>"
+
+  - name: qwen/qwq-32b-preview
+    tokenizer_spec:
+      class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"
+    end_of_text_token: "<|im_end|>"
+    prefix_token: ""
 
   - name: qwen/qwen-vl
     tokenizer_spec: