mirror of
https://github.com/infiniflow/ragflow.git
synced 2026-07-04 18:45:38 +08:00
d8d49df35e
### What problem does this PR solve?
The Gitee AI Go driver shipped with a stub \`Rerank\` method that
returned \`Rerank not implemented\`, even though
\`conf/models/gitee.json\` already wires the rerank URL suffix at
\`\"rerank\": \"rerank\"\`. The same config did not list any
rerank model, so the picker had nothing to select.
So a Gitee tenant could not use BAAI/bge-reranker-v2-m3 as a reranker
through the Go layer today, even though the
infrastructure was one config entry and one method body away.
### What this PR includes
- \`conf/models/gitee.json\`: add \`BAAI/bge-reranker-v2-m3\` to the
\`models\` array.
- \`internal/entity/models/gitee.go\`: replace the \`Rerank\` stub with
a real implementation. Adds two small local types
that match the OpenAI-compatible \`/rerank\` shape already used by the
SiliconFlow and ZhipuAI drivers.
No factory change. No interface change.
### How the driver works
- Validate \`apiConfig\` and the API key, validate the model name,
resolve the region with a default fallback, build the
URL from \`BaseURL[region] + URLSuffix.Rerank\`.
- Use a per-call \`context.WithTimeout(30s)\` and
\`http.NewRequestWithContext\`, matching the pattern the
recently merged Aliyun Encode and the OpenAI driver already use.
- Send \`{model, query, documents, top_n, return_documents:false}\` in
the body.
- Parse \`results[*].relevance_score\` and copy each score into the
output slice indexed by \`results[*].index\`, so the
output order matches the input order even if the API returns items in a
different order.
- Empty input returns \`[]float64{}\` with no HTTP call.
- An out-of-range result index returns a clear error rather than
silently skipping the entry.
- Non-200 responses propagate the upstream status line and body.
### Type of change
- [x] New Feature (non-breaking change which adds functionality)
### How was this tested?
- \`go build ./internal/entity/models/...\` in a clean go 1.25 image
returns exit 0.
- The full method set on \`GiteeModel\` still matches the
\`ModelDriver\` interface.
- Pattern parity with the existing SiliconFlow Rerank and the recently
merged ZhipuAI Rerank (#14608).
Closes #14655
44 lines
761 B
JSON
44 lines
761 B
JSON
{
|
|
"name": "Gitee",
|
|
"url": {
|
|
"default": "https://api.moark.com/v1"
|
|
},
|
|
"url_suffix": {
|
|
"chat": "chat/completions",
|
|
"models": "models",
|
|
"status": "",
|
|
"balance": "tokens/packages/balance",
|
|
"embedding": "embedding",
|
|
"rerank": "rerank"
|
|
},
|
|
"models": [
|
|
{
|
|
"name": "qwen3-8b",
|
|
"max_tokens": 32768,
|
|
"model_types": [
|
|
"chat"
|
|
]
|
|
},
|
|
{
|
|
"name": "qwen3-0.6b",
|
|
"max_tokens": 32768,
|
|
"model_types": [
|
|
"chat"
|
|
]
|
|
},
|
|
{
|
|
"name": "glm-4.7-flash",
|
|
"max_tokens": 204800,
|
|
"model_types": [
|
|
"chat"
|
|
]
|
|
},
|
|
{
|
|
"name": "BAAI/bge-reranker-v2-m3",
|
|
"max_tokens": 8192,
|
|
"model_types": [
|
|
"rerank"
|
|
]
|
|
}
|
|
]
|
|
} |