adding zephyr 7b (#389)

* adding zephyr 7b * update tokenizer repo
scaleapi · Nov 27, 2023 · 37814ee · 37814ee
1 parent 3c0f168
commit 37814ee
Show file tree

Hide file tree

Showing 3 changed files with 8 additions and 0 deletions.
diff --git a/docs/model_zoo.md b/docs/model_zoo.md
@@ -26,6 +26,8 @@ Scale hosts the following models in the LLM Engine Model Zoo:
 | `codellama-13b-instruct`         | ✅                       | ✅                           | text-generation-inference, vllm |
 | `codellama-34b`         | ✅                       | ✅                           | text-generation-inference, vllm |
 | `codellama-34b-instruct`         | ✅                       | ✅                           | text-generation-inference, vllm |
+| `zephyr-7b-alpha`         | ✅                       |                            | text-generation-inference, vllm |
+| `zephyr-7b-beta`         | ✅                       |                            | text-generation-inference, vllm |
 
 ## Usage
 

diff --git a/model-engine/model_engine_server/domain/use_cases/llm_model_endpoint_use_cases.py b/model-engine/model_engine_server/domain/use_cases/llm_model_endpoint_use_cases.py
@@ -124,6 +124,8 @@
             "codellama-34b-instruct",
             "llm-jp-13b-instruct-full",
             "llm-jp-13b-instruct-full-dolly",
+            "zephyr-7b-alpha",
+            "zephyr-7b-beta",
         ]
     ),
     LLMInferenceFramework.VLLM: set(
@@ -154,6 +156,8 @@
             "mammoth-coder-llama-2-7b",
             "mammoth-coder-llama-2-13b",
             "mammoth-coder-llama-2-34b",
+            "zephyr-7b-alpha",
+            "zephyr-7b-beta",
         ]
     ),
     LLMInferenceFramework.LIGHTLLM: set(

diff --git a/model-engine/model_engine_server/infra/repositories/live_tokenizer_repository.py b/model-engine/model_engine_server/infra/repositories/live_tokenizer_repository.py
@@ -67,6 +67,8 @@ def get_default_supported_models_info() -> Dict[str, ModelInfo]:
         "dolly-v2-12b": ModelInfo("databricks/dolly-v2-12b", None),
         "stablelm-tuned-7b": ModelInfo("StabilityAI/stablelm-tuned-alpha-7b", None),
         "vicuna-13b": ModelInfo("eachadea/vicuna-13b-1.1", None),
+        "zephyr-7b-alpha": ModelInfo("HuggingFaceH4/zephyr-7b-alpha", None),
+        "zephyr-7b-beta": ModelInfo("HuggingFaceH4/zephyr-7b-beta", None),
     }