Granite-4.0-H-Micro
- Tool use
- Open weights
Prices
0 offerings · prices as listedNo priced offering is listed for this model yet.
Details
- Licence
- apache-2.0
- Family
- granite-h-micro
Not stated on IBM’s pages: max input, input modalities, output modalities, reasoning, reasoning efforts, reasoning can be off, max thinking budget, structured output, built-in tools, knowledge cutoff, aliases.
Run locally
Weights: ibm-granite/granite-4.0-h-micro- Parameters 3.2B
- Dense
- Native precision BF16
- Layers 40
- KV cache per token 8 KB
- Config positions 128K (config.json max_position_embeddings)
- Needs
- ≈ 6.5 GB (weights 5.9 (maker’s file) + KV cache 0.1 + overhead 0.5)
- Usable
- 21.6 GB (90% of its memory)
- Generation speed
- 78–117 tok/s
- Longest context
- 128K (the model’s limit)
- First token
- about 0.10–0.15 s for a 2K prompt
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| BF16 (native)maker’s file | 5.9 GB | ≈ 6.5 | ●Runs well 78–117 tok/s | 128K |
| Q8_0maker’s file | 3.2 GB | ≈ 3.7 | ●Runs well 145–218 tok/s | 128K |
| FP8 | 3.0 GB | ≈ 3.5 | ●Runs well 155–232 tok/s | 128K |
| Q6_Kmaker’s file | 2.4 GB | ≈ 3.0 | ●Runs well 187–281 tok/s | 128K |
| Q5_K_Mmaker’s file | 2.1 GB | ≈ 2.7 | ●Runs well 215–323 tok/s | 128K |
| Q5_0maker’s file | 2.1 GB | ≈ 2.6 | ●Runs well 220–329 tok/s | 128K |
| Q4_K_Mmaker’s file | 1.8 GB | ≈ 2.4 | ●Runs well 251–376 tok/s | 128K |
| Q4_0maker’s file | 1.7 GB | ≈ 2.3 | ●Runs well 262–393 tok/s | 128K |
| IQ4_XS | 1.6 GB | ≈ 2.1 | ●Runs well 286–429 tok/s | 128K |
| Q3_K_Mmaker’s filequality drops | 1.4 GB | ≈ 2.0 | ●Runs well 311–466 tok/s | 128K |
| IQ3_XXSquality drops | 1.1 GB | ≈ 1.7 | ●Runs well 391–587 tok/s | 128K |
| Q2_Kmaker’s filequality drops | 1.1 GB | ≈ 1.7 | ●Runs well 390–585 tok/s | 128K |
| IQ2_XXSquality drops | 0.8 GB | ≈ 1.3 | ●Runs well 567–850 tok/s | 128K |
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 ibm-granite/granite-4.0-h-micro --max-model-len 8192
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 usThis model isn’t on any leaderboard we cite yet.
Changes
All changes →No changes recorded yet.
Sources
- ibm-granite/granite-4.0-h-micro model card (opens in a new tab) huggingface.co · checked Nov 2, 2026
- ibm-granite/granite-4.0-h-micro model card (opens in a new tab) huggingface.co · checked Nov 4, 2026
- IBM foundation models (opens in a new tab) ibm.com · checked Oct 31, 2026
- ibm-granite/granite-4.0-h-micro config.json (opens in a new tab) huggingface.co · checked Oct 30, 2026
- ibm-granite/granite-4.0-h-micro files (opens in a new tab) huggingface.co · checked Nov 1, 2026
- ibm-granite/granite-4.0-h-micro-GGUF files (opens in a new tab) huggingface.co · checked Nov 2, 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
IBM · United States · ibm.com
Related models
Snapshots, the same family and other models from this provider