diff --git a/extensions/inference-nitro-extension/resources/models/qwen2.5-14b-instruct/model.json b/extensions/inference-nitro-extension/resources/models/qwen2.5-14b-instruct/model.json index f194b3f16..04913b874 100644 --- a/extensions/inference-nitro-extension/resources/models/qwen2.5-14b-instruct/model.json +++ b/extensions/inference-nitro-extension/resources/models/qwen2.5-14b-instruct/model.json @@ -22,7 +22,7 @@ "top_p": 0.95, "stream": true, "max_tokens": 32768, - "stop": [], + "stop": ["<|endoftext|>", "<|im_end|>"], "frequency_penalty": 0, "presence_penalty": 0 }, diff --git a/extensions/inference-nitro-extension/resources/models/qwen2.5-32b-instruct/model.json b/extensions/inference-nitro-extension/resources/models/qwen2.5-32b-instruct/model.json index 7b9b1ab88..43ba30c56 100644 --- a/extensions/inference-nitro-extension/resources/models/qwen2.5-32b-instruct/model.json +++ b/extensions/inference-nitro-extension/resources/models/qwen2.5-32b-instruct/model.json @@ -22,7 +22,7 @@ "top_p": 0.95, "stream": true, "max_tokens": 32768, - "stop": [], + "stop": ["<|endoftext|>", "<|im_end|>"], "frequency_penalty": 0, "presence_penalty": 0 }, diff --git a/extensions/inference-nitro-extension/resources/models/qwen2.5-72b-instruct/model.json b/extensions/inference-nitro-extension/resources/models/qwen2.5-72b-instruct/model.json index 73e000406..1852a0909 100644 --- a/extensions/inference-nitro-extension/resources/models/qwen2.5-72b-instruct/model.json +++ b/extensions/inference-nitro-extension/resources/models/qwen2.5-72b-instruct/model.json @@ -22,7 +22,7 @@ "top_p": 0.95, "stream": true, "max_tokens": 32768, - "stop": [], + "stop": ["<|endoftext|>", "<|im_end|>"], "frequency_penalty": 0, "presence_penalty": 0 }, diff --git a/extensions/inference-nitro-extension/resources/models/qwen2.5-7b-instruct/model.json b/extensions/inference-nitro-extension/resources/models/qwen2.5-7b-instruct/model.json index 2d0112bdb..b47511f96 100644 --- a/extensions/inference-nitro-extension/resources/models/qwen2.5-7b-instruct/model.json +++ b/extensions/inference-nitro-extension/resources/models/qwen2.5-7b-instruct/model.json @@ -22,7 +22,7 @@ "top_p": 0.95, "stream": true, "max_tokens": 32768, - "stop": [], + "stop": ["<|endoftext|>", "<|im_end|>"], "frequency_penalty": 0, "presence_penalty": 0 }, diff --git a/extensions/inference-nitro-extension/resources/models/qwen2.5-coder-7b-instruct/model.json b/extensions/inference-nitro-extension/resources/models/qwen2.5-coder-7b-instruct/model.json index 23b58d163..2f1080b2c 100644 --- a/extensions/inference-nitro-extension/resources/models/qwen2.5-coder-7b-instruct/model.json +++ b/extensions/inference-nitro-extension/resources/models/qwen2.5-coder-7b-instruct/model.json @@ -22,7 +22,7 @@ "top_p": 0.95, "stream": true, "max_tokens": 32768, - "stop": [], + "stop": ["<|endoftext|>", "<|im_end|>"], "frequency_penalty": 0, "presence_penalty": 0 },