diff --git a/assets/models/system/Llama-2-13b-chat/spec.yaml b/assets/models/system/Llama-2-13b-chat/spec.yaml index de4ccf795c..b082ae1762 100644 --- a/assets/models/system/Llama-2-13b-chat/spec.yaml +++ b/assets/models/system/Llama-2-13b-chat/spec.yaml @@ -43,5 +43,6 @@ tags: inference_supported_envs: - vllm - ds_mii + benchmark: "quality" version: 20 diff --git a/assets/models/system/Llama-2-13b/spec.yaml b/assets/models/system/Llama-2-13b/spec.yaml index 9cdcc5f013..8e1b8e91d1 100644 --- a/assets/models/system/Llama-2-13b/spec.yaml +++ b/assets/models/system/Llama-2-13b/spec.yaml @@ -61,5 +61,6 @@ tags: inference_supported_envs: - vllm - ds_mii + benchmark: "quality" version: 23 diff --git a/assets/models/system/Llama-2-70b-chat/spec.yaml b/assets/models/system/Llama-2-70b-chat/spec.yaml index 0c41284146..365f162604 100644 --- a/assets/models/system/Llama-2-70b-chat/spec.yaml +++ b/assets/models/system/Llama-2-70b-chat/spec.yaml @@ -35,5 +35,6 @@ tags: - Standard_ND96asr_v4 - Standard_ND96amsr_A100_v4 - Standard_NC96ads_A100_v4 + benchmark: "quality" version: 20 diff --git a/assets/models/system/Llama-2-70b/spec.yaml b/assets/models/system/Llama-2-70b/spec.yaml index 6182588dc5..5ee52c0525 100644 --- a/assets/models/system/Llama-2-70b/spec.yaml +++ b/assets/models/system/Llama-2-70b/spec.yaml @@ -49,5 +49,6 @@ tags: deepspeed_stage: "3" precision: "16" ignore_mismatched_sizes: "false" + benchmark: "quality" version: 24 diff --git a/assets/models/system/Llama-2-7b-chat/spec.yaml b/assets/models/system/Llama-2-7b-chat/spec.yaml index eb6ebe1fa4..5cacdfc71e 100644 --- a/assets/models/system/Llama-2-7b-chat/spec.yaml +++ b/assets/models/system/Llama-2-7b-chat/spec.yaml @@ -58,5 +58,6 @@ tags: inference_supported_envs: - vllm - ds_mii + benchmark: "quality" version: 24 diff --git a/assets/models/system/Llama-2-7b/spec.yaml b/assets/models/system/Llama-2-7b/spec.yaml index f6b96ec398..9507aff18c 100644 --- a/assets/models/system/Llama-2-7b/spec.yaml +++ b/assets/models/system/Llama-2-7b/spec.yaml @@ -61,5 +61,6 @@ tags: inference_supported_envs: - vllm - ds_mii + benchmark: "quality" version: 22 diff --git a/assets/models/system/Mistral-7B-Instruct-v0-1/spec.yaml b/assets/models/system/Mistral-7B-Instruct-v0-1/spec.yaml index 8cbd7cc82a..c38896ef6b 100644 --- a/assets/models/system/Mistral-7B-Instruct-v0-1/spec.yaml +++ b/assets/models/system/Mistral-7B-Instruct-v0-1/spec.yaml @@ -30,4 +30,5 @@ tags: license: apache-2.0 task: chat-completion author: "Mistral AI" + benchmark: "quality" version: 10 diff --git a/assets/models/system/Mistral-7B-Instruct-v0-2/spec.yaml b/assets/models/system/Mistral-7B-Instruct-v0-2/spec.yaml index b9493f8b9a..88d8e4198d 100644 --- a/assets/models/system/Mistral-7B-Instruct-v0-2/spec.yaml +++ b/assets/models/system/Mistral-7B-Instruct-v0-2/spec.yaml @@ -29,4 +29,5 @@ tags: license: apache-2.0 task: chat-completion author: "Mistral AI" + benchmark: "quality" version: 5 diff --git a/assets/models/system/Mistral-7B-v0-1/spec.yaml b/assets/models/system/Mistral-7B-v0-1/spec.yaml index 9c4d4b5495..1defd10004 100644 --- a/assets/models/system/Mistral-7B-v0-1/spec.yaml +++ b/assets/models/system/Mistral-7B-v0-1/spec.yaml @@ -71,4 +71,5 @@ tags: license: apache-2.0 task: text-generation author: "Mistral AI" + benchmark: "quality" version: 17 diff --git a/assets/models/system/Mixtral-8x7B-v0-1/spec.yaml b/assets/models/system/Mixtral-8x7B-v0-1/spec.yaml index 5aa9e02108..3a5d7a9ee2 100644 --- a/assets/models/system/Mixtral-8x7B-v0-1/spec.yaml +++ b/assets/models/system/Mixtral-8x7B-v0-1/spec.yaml @@ -41,4 +41,5 @@ tags: license: apache-2.0 task: text-generation author: "Mistral AI" + benchmark: "quality" version: 14 diff --git a/assets/models/system/microsoft-phi-2/spec.yaml b/assets/models/system/microsoft-phi-2/spec.yaml index 29e59f8798..e4f5661ea5 100644 --- a/assets/models/system/microsoft-phi-2/spec.yaml +++ b/assets/models/system/microsoft-phi-2/spec.yaml @@ -60,4 +60,5 @@ tags: apply_ort: "false" precision: 16 max_seq_length: 2048 + benchmark: "quality" version: 19 diff --git a/assets/models/system/mistral-community-Mixtral-8x22B-v01/spec.yaml b/assets/models/system/mistral-community-Mixtral-8x22B-v01/spec.yaml index 96c02c3652..b3ce01f371 100644 --- a/assets/models/system/mistral-community-Mixtral-8x22B-v01/spec.yaml +++ b/assets/models/system/mistral-community-Mixtral-8x22B-v01/spec.yaml @@ -20,4 +20,5 @@ tags: license: apache-2.0 task: text-generation author: "Mistral AI" + benchmark: "quality" version: 5 diff --git a/assets/models/system/mistralai-Mixtral-8x22B-Instruct-v0-1/spec.yaml b/assets/models/system/mistralai-Mixtral-8x22B-Instruct-v0-1/spec.yaml index 89e97bfe97..2ffb43af3b 100644 --- a/assets/models/system/mistralai-Mixtral-8x22B-Instruct-v0-1/spec.yaml +++ b/assets/models/system/mistralai-Mixtral-8x22B-Instruct-v0-1/spec.yaml @@ -20,4 +20,5 @@ tags: license: apache-2.0 task: chat-completion author: "Mistral AI" + benchmark: "quality" version: 4 diff --git a/assets/models/system/mistralai-Mixtral-8x22B-v0-1/spec.yaml b/assets/models/system/mistralai-Mixtral-8x22B-v0-1/spec.yaml index ce8a3fb018..9253984a02 100644 --- a/assets/models/system/mistralai-Mixtral-8x22B-v0-1/spec.yaml +++ b/assets/models/system/mistralai-Mixtral-8x22B-v0-1/spec.yaml @@ -20,4 +20,5 @@ tags: license: apache-2.0 task: text-generation author: "Mistral AI" + benchmark: "quality" version: 4 diff --git a/assets/models/system/mistralai-Mixtral-8x7B-Instruct-v01/spec.yaml b/assets/models/system/mistralai-Mixtral-8x7B-Instruct-v01/spec.yaml index c7b83eea36..c8c164bf40 100644 --- a/assets/models/system/mistralai-Mixtral-8x7B-Instruct-v01/spec.yaml +++ b/assets/models/system/mistralai-Mixtral-8x7B-Instruct-v01/spec.yaml @@ -26,4 +26,5 @@ tags: license: apache-2.0 task: chat-completion author: "Mistral AI" + benchmark: "quality" version: 9 diff --git a/assets/models/system/phi-3-medium-128k-instruct/spec.yaml b/assets/models/system/phi-3-medium-128k-instruct/spec.yaml index 8fd6c4f709..da485b1a46 100644 --- a/assets/models/system/phi-3-medium-128k-instruct/spec.yaml +++ b/assets/models/system/phi-3-medium-128k-instruct/spec.yaml @@ -68,4 +68,5 @@ tags: logging_steps: 10 save_total_limit: 1 max_seq_length: 4096 + benchmark: "quality" version: 4 diff --git a/assets/models/system/phi-3-medium-4k-instruct/spec.yaml b/assets/models/system/phi-3-medium-4k-instruct/spec.yaml index 99491d570b..b3dcd36172 100644 --- a/assets/models/system/phi-3-medium-4k-instruct/spec.yaml +++ b/assets/models/system/phi-3-medium-4k-instruct/spec.yaml @@ -71,4 +71,5 @@ tags: logging_steps: 10 save_total_limit: 1 max_seq_length: 4096 + benchmark: "quality" version: 4 diff --git a/assets/models/system/phi-3-mini-128k-instruct/spec.yaml b/assets/models/system/phi-3-mini-128k-instruct/spec.yaml index e2eb1cf358..65ee2d6e65 100644 --- a/assets/models/system/phi-3-mini-128k-instruct/spec.yaml +++ b/assets/models/system/phi-3-mini-128k-instruct/spec.yaml @@ -70,4 +70,5 @@ tags: logging_strategy: "steps" logging_steps: 10 save_total_limit: 1 + benchmark: "quality" version: 11 diff --git a/assets/models/system/phi-3-mini-4k-instruct/spec.yaml b/assets/models/system/phi-3-mini-4k-instruct/spec.yaml index 8ba2f62079..f0abda69e2 100644 --- a/assets/models/system/phi-3-mini-4k-instruct/spec.yaml +++ b/assets/models/system/phi-3-mini-4k-instruct/spec.yaml @@ -76,4 +76,5 @@ tags: logging_strategy: "steps" logging_steps: 10 save_total_limit: 1 + benchmark: "quality" version: 11 diff --git a/assets/models/system/phi-3-small-128k-instruct/spec.yaml b/assets/models/system/phi-3-small-128k-instruct/spec.yaml index 26997494b6..02b1cfa419 100644 --- a/assets/models/system/phi-3-small-128k-instruct/spec.yaml +++ b/assets/models/system/phi-3-small-128k-instruct/spec.yaml @@ -64,4 +64,5 @@ tags: logging_steps: 10 save_total_limit: 1 max_seq_length: 4096 + benchmark: "quality" version: 4 diff --git a/assets/models/system/phi-3-small-8k-instruct/spec.yaml b/assets/models/system/phi-3-small-8k-instruct/spec.yaml index 4fb0c9c694..4b47cc634b 100644 --- a/assets/models/system/phi-3-small-8k-instruct/spec.yaml +++ b/assets/models/system/phi-3-small-8k-instruct/spec.yaml @@ -64,4 +64,5 @@ tags: logging_steps: 10 save_total_limit: 1 max_seq_length: 4096 + benchmark: "quality" version: 4 diff --git a/assets/models/system/stabilityai-stable-diffusion-2-1/spec.yaml b/assets/models/system/stabilityai-stable-diffusion-2-1/spec.yaml index 2bc47f7202..5343208fae 100644 --- a/assets/models/system/stabilityai-stable-diffusion-2-1/spec.yaml +++ b/assets/models/system/stabilityai-stable-diffusion-2-1/spec.yaml @@ -40,4 +40,5 @@ tags: Standard_ND96amsr_A100_v4, Standard_ND96asr_v4 ] + benchmark: "quality" version: 12