Add GPT-4o model (#2649)

Co-authored-by: Tony Lee <[email protected]>
stanford-crfm · May 15, 2024 · 5908b07 · 5908b07
1 parent 05b44e3
commit 5908b07
Show file tree

Hide file tree

Showing 4 changed files with 22 additions and 1 deletion.
diff --git a/setup.cfg b/setup.cfg
@@ -134,7 +134,7 @@ mistral =
 
 openai =
     openai~=1.0
-    tiktoken~=0.3.3
+    tiktoken~=0.7
     pydantic~=2.0  # For model_dump(mode="json") - openai only requires pydantic>=1.9.0
 
 google =

diff --git a/src/helm/config/model_deployments.yaml b/src/helm/config/model_deployments.yaml
@@ -1348,6 +1348,13 @@ model_deployments:
     client_spec:
       class_name: "helm.clients.openai_client.OpenAIClient"
 
+  - name: openai/gpt-4o-2024-05-13
+    model_name: openai/gpt-4o-2024-05-13
+    tokenizer_name: openai/o200k_base
+    max_sequence_length: 128000
+    client_spec:
+      class_name: "helm.clients.openai_client.OpenAIClient"
+
   - name: openai/gpt-4-vision-preview
     model_name: openai/gpt-4-vision-preview
     tokenizer_name: openai/cl100k_base

diff --git a/src/helm/config/model_metadata.yaml b/src/helm/config/model_metadata.yaml
@@ -1806,6 +1806,14 @@ models:
     release_date: 2024-04-09
     tags: [TEXT_MODEL_TAG, OPENAI_CHATGPT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG, INSTRUCTION_FOLLOWING_MODEL_TAG]
 
+  - name: openai/gpt-4o-2024-05-13
+    display_name: GPT-4o (2024-05-13)
+    description: GPT-4o (2024-05-13) is a large multimodal model that accepts as input any combination of text, audio, and image and generates any combination of text, audio, and image outputs.
+    creator_organization_name: OpenAI
+    access: limited
+    release_date: 2024-04-09
+    tags: [TEXT_MODEL_TAG, VISION_LANGUAGE_MODEL_TAG, OPENAI_CHATGPT_MODEL_TAG, LIMITED_FUNCTIONALITY_TEXT_MODEL_TAG, INSTRUCTION_FOLLOWING_MODEL_TAG]
+
   - name: openai/gpt-4-vision-preview
     # According to https://platform.openai.com/docs/models/gpt-4-turbo-and-gpt-4, this model has pointed gpt-4-1106-vision-preview.
     display_name: GPT-4V (1106 preview)

diff --git a/src/helm/config/tokenizer_configs.yaml b/src/helm/config/tokenizer_configs.yaml
@@ -288,6 +288,12 @@ tokenizer_configs:
     end_of_text_token: "<|endoftext|>"
     prefix_token: "<|endoftext|>"
 
+  - name: openai/o200k_base
+    tokenizer_spec:
+      class_name: "helm.tokenizers.tiktoken_tokenizer.TiktokenTokenizer"
+    end_of_text_token: "<|endoftext|>"
+    prefix_token: "<|endoftext|>"
+
   - name: openai/clip-vit-large-patch14
     tokenizer_spec:
       class_name: "helm.tokenizers.huggingface_tokenizer.HuggingFaceTokenizer"