bge-reranker-base
- Open weights
Prices
0 offerings · prices as listedNo priced offering is listed for this model yet.
Details
- Languages
- 2
Not stated on BAAI’s pages: input modalities.
Run locally
Weights: BAAI/bge-reranker-base- Parameters 0.3B
- Dense
- Native precision FP32
- Layers 12
- KV cache per token 36 KB
- Config positions 514 (config.json max_position_embeddings)
- Needs
- ≈ 1.6 GB (weights 1.0 (maker’s file) + KV cache 0.0 + overhead 0.5)
- Usable
- 21.6 GB (90% of its memory)
- Generation speed
- 446–668 tok/s
- Longest context
- 514 (the model’s limit)
- First token
- about 0.01–0.01 s for a 2K prompt
| Precision | Weights | Needs at 514 | On your hardware | Longest context |
|---|---|---|---|---|
| FP32 (native)maker’s file | 1.0 GB | ≈ 1.6 | ●Runs well 446–668 tok/s | 514 |
| Q8_0 | 0.3 GB | ≈ 0.8 | ●Runs well 1603–2405 tok/s | 514 |
| FP8 | 0.3 GB | ≈ 0.8 | ●Runs well 1697–2546 tok/s | 514 |
| Q6_K | 0.2 GB | ≈ 0.7 | ●Runs well 2038–3057 tok/s | 514 |
| Q5_K_M | 0.2 GB | ≈ 0.7 | ●Runs well 2326–3489 tok/s | 514 |
| Q5_0 | 0.2 GB | ≈ 0.7 | ●Runs well 2399–3598 tok/s | 514 |
| Q4_K_M | 0.2 GB | ≈ 0.7 | ●Runs well 2683–4024 tok/s | 514 |
| Q4_0 | 0.1 GB | ≈ 0.7 | ●Runs well 2874–4311 tok/s | 514 |
| IQ4_XS | 0.1 GB | ≈ 0.7 | ●Runs well 3024–4536 tok/s | 514 |
| Q3_K_Mquality drops | 0.1 GB | ≈ 0.6 | ●Runs well 3255–4882 tok/s | 514 |
| IQ3_XXSquality drops | 0.1 GB | ≈ 0.6 | ●Runs well 4022–6034 tok/s | 514 |
| Q2_Kquality drops | 0.1 GB | ≈ 0.6 | ●Runs well 4090–6136 tok/s | 514 |
| IQ2_XXSquality drops | 0.1 GB | ≈ 0.6 | ●Runs well 5566–8349 tok/s | 514 |
Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.
Compared with other hardware
Each at the highest precision it runs. Click one for details.
Smallest that runs it (Q4_K_M, 8K): GPU GeForce RTX 3080; unified memory Apple M5 · 16 GB.
Run commands
vllm serve BAAI/bge-reranker-base --max-model-len 514
Memory to fine-tune
Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.
Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated
Cited scores
From third-party leaderboards, not measured by usNo cited leaderboard covers rerank models yet.
Changes
All changes →No changes recorded yet.
Sources
- BAAI/bge-reranker-base model card (opens in a new tab) huggingface.co · checked Oct 31, 2026
- BAAI/bge-reranker-base model card (opens in a new tab) huggingface.co · checked Nov 3, 2026
- BAAI/bge-reranker-base config.json (opens in a new tab) huggingface.co · checked Nov 1, 2026
- BAAI/bge-reranker-base files (opens in a new tab) huggingface.co · checked Nov 2, 2026
- FlagEmbedding (opens in a new tab) huggingface.co · checked Oct 30, 2026
Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data
BAAI · China · baai.ac.cn
Related models
Snapshots, the same family and other models from this provider