[Model] Add zerank-2 reranker (#60415)

Signed-off-by: rdlhio <contact@rdlh.io>
This commit is contained in:
Rémi Delhaye
2026-10-09 03:01:51 +00:00
committed by GitHub
parent 702ce313f4
commit a1cd0ec650
5 changed files with 119 additions and 1 deletions
@@ -0,0 +1,13 @@
{#- zerank-2 (zeroentropy/zerank-2-reranker): query in the system turn, document in the user turn.
The assistant suffix is emitted unconditionally, since score templates are rendered
without add_generation_prompt. An optional `instruction` is wrapped as recommended by
ZeroEntropy; without it, the prompt is the official zerank-2 prompt. -#}
{%- set query = messages | selectattr("role", "eq", "query") | map(attribute="content") | list -%}
{%- set document = messages | selectattr("role", "eq", "document") | map(attribute="content") | list -%}
{%- set query = query[0] if query else '' -%}
{%- if instruction is defined and instruction is string and instruction | trim -%}
{%- set query = '<query>' + query + '</query>\n<instruction>' + instruction + '</instruction>' -%}
{%- endif -%}
{{- '<|im_start|>system\n' + query + '<|im_end|>\n' -}}
{{- '<|im_start|>user\n' + (document[0] if document else '') + '<|im_end|>\n' -}}
{{- '<|im_start|>assistant\n' -}}
@@ -53,6 +53,7 @@ def get_chat_template(model: str) -> str:
"mixedbread-ai/mxbai-rerank-base-v2": "mxbai_rerank_v2.jinja",
"mixedbread-ai/mxbai-rerank-large-v2": "mxbai_rerank_v2.jinja",
"nvidia/llama-nemotron-rerank-1b-v2": "nemotron-rerank.jinja",
"polaria-tech/zerank-2-reranker-vllm": "zerank2.jinja",
}
# Get the template filename for the specified model
@@ -110,6 +111,7 @@ def get_hf_overrides(model: str) -> dict[str, Any]:
"method": "from_2_way_softmax",
},
"nvidia/llama-nemotron-rerank-1b-v2": {},
"polaria-tech/zerank-2-reranker-vllm": {},
}
hf_overrides = model_name_to_hf_overrides_map.get(model)
@@ -19,6 +19,7 @@ run:
vllm serve tomaarsen/Qwen3-Reranker-0.6B-seq-cls --chat-template examples/pooling/score/template/qwen3_reranker.jinja
vllm serve mixedbread-ai/mxbai-rerank-base-v2 --hf_overrides '{"architectures": ["Qwen2ForSequenceClassification"],"classifier_from_token": ["0", "1"], "method": "from_2_way_softmax"}' --chat-template examples/pooling/score/template/mxbai_rerank_v2.jinja
vllm serve nvidia/llama-nemotron-rerank-1b-v2 --runner pooling --trust-remote-code --chat-template examples/pooling/score/template/nemotron-rerank.jinja
vllm serve polaria-tech/zerank-2-reranker-vllm --runner pooling --chat-template examples/pooling/score/template/zerank2.jinja
vllm serve Qwen/Qwen3-Reranker-0.6B --runner pooling --hf_overrides '{"architectures": ["Qwen3ForSequenceClassification"],"classifier_from_token": ["no", "yes"],"is_original_qwen3_reranker": true}' --chat-template examples/pooling/score/template/qwen3_reranker.jinja
"""