From 8232a1e0c2bbc5a83ed8f35bfa4a924dba17c938 Mon Sep 17 00:00:00 2001 From: Van Pham <64197333+Van-QA@users.noreply.github.com> Date: Wed, 28 Aug 2024 10:44:24 +0700 Subject: [PATCH 1/2] Update model.json --- .../resources/models/gemma-1.1-2b/model.json | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/extensions/inference-nitro-extension/resources/models/gemma-1.1-2b/model.json b/extensions/inference-nitro-extension/resources/models/gemma-1.1-2b/model.json index 56cd9c81cb..837b10ce3a 100644 --- a/extensions/inference-nitro-extension/resources/models/gemma-1.1-2b/model.json +++ b/extensions/inference-nitro-extension/resources/models/gemma-1.1-2b/model.json @@ -1,7 +1,7 @@ { "sources": [ { - "filename": "gemma-1.1-2b-it-q4_k_m.gguf", + "filename": "gemma-1.1-2b-it-Q4_K_M.gguf", "url": "https://huggingface.co/bartowski/gemma-1.1-2b-it-GGUF/resolve/main/gemma-1.1-2b-it-Q4_K_M.gguf" } ], From 47b074af471bbcb620b2e80c0c2ba2040a070370 Mon Sep 17 00:00:00 2001 From: Van Pham <64197333+Van-QA@users.noreply.github.com> Date: Wed, 28 Aug 2024 10:46:20 +0700 Subject: [PATCH 2/2] Update model.json --- .../resources/models/gemma-1.1-7b/model.json | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/extensions/inference-nitro-extension/resources/models/gemma-1.1-7b/model.json b/extensions/inference-nitro-extension/resources/models/gemma-1.1-7b/model.json index 5bd89b478a..a532c1dc35 100644 --- a/extensions/inference-nitro-extension/resources/models/gemma-1.1-7b/model.json +++ b/extensions/inference-nitro-extension/resources/models/gemma-1.1-7b/model.json @@ -1,7 +1,7 @@ { "sources": [ { - "filename": "gemma-1.1-7b-it-q4_K_M.gguf", + "filename": "gemma-1.1-7b-it-Q4_K_M.gguf", "url": "https://huggingface.co/bartowski/gemma-1.1-7b-it-GGUF/resolve/main/gemma-1.1-7b-it-Q4_K_M.gguf" } ], @@ -14,7 +14,7 @@ "settings": { "ctx_len": 8192, "prompt_template": "user\n{prompt}\nmodel", - "llama_model_path": "gemma-1.1-7b-it-q4_K_M.gguf", + "llama_model_path": "gemma-1.1-7b-it-Q4_K_M.gguf", "ngl": 29 }, "parameters": {