ragflow

mirror of https://github.com/infiniflow/ragflow.git synced 2026-08-08 00:18:12 +08:00

Files

jay77721 6be7dc6e66 feat: add batch_size to embedding models in all_models.json (#17878 )

## Summary

Add a `batch_size` field to all **215 embedding models** in
`conf/all_models.json`, synced from the values set in
`conf/models/*.json` (see #17877) so the two config sources stay
consistent.

## Matching strategy

Values are assigned by layered matching against `conf/models/*.json`:

1. **Exact model name** match
2. **Alias** match
3. **Bare-name** map (entries that appear without a provider prefix)
4. **Provider-prefix family** match (e.g. `jinaai/*`, `nvidia/*`,
`cohere/*`, `qwen/*`, `BAAI/*`, `voyage/*`, `zai-org/*`)

## Distribution

| batch_size | # models | Source |
|---|---|---|
| 16384 | 88 | Jina (`jinaai/*`) — no documented batch limit, safe high
cap |
| 2048 | 26 | OpenAI + OpenAI-compatible proxies (Qwen, BGE, e5
families) |
| 32 | 26 | NVIDIA NIM, BAAI/bge-m3, qwen3-embedding |
| 1 | 27 | Conservative provider fallback where no API batch limit is
documented |
| 1000 | 24 | Voyage AI — API reference max |
| 96 | 8 | Cohere |
| 16 | 5 | Mistral, Baidu Qianfan, Baichuan, Replicate |
| 100 | 4 | Google Gemini |
| 512 | 3 | Zhipu GLM (`zai-org/embedding-2/3`, `zhipu-embedding-2`) |
| 10 | 3 | Volcengine Doubao |
| 1024 | 1 | DeepInfra Qwen3-Embedding-4B |

2026-08-06 10:48:05 +08:00

models

feat: add batch_size to all embedding model configs (#17877 )

2026-08-06 10:47:38 +08:00

all_models.json

feat: add batch_size to embedding models in all_models.json (#17878 )

2026-08-06 10:48:05 +08:00

doc_meta_es_mapping.json

Put document metadata in ES/Infinity (#12826 )

2026-01-28 13:29:34 +08:00

doc_meta_infinity_mapping.json

Put document metadata in ES/Infinity (#12826 )