MiniCPM-V 4.7 1B (MNN)
The language model of OpenBMB's MiniCPM-V 4.7 1B as a 4-bit MNN bundle for text chat on your phone with TokForge.
Published by TokForge for on-device inference.
| What it is | Chat model (text only; the vision part is not included) |
| Format | MNN bundle, 4-bit HQQ weights (block 64), BF16 embeddings |
| Runs on | Android phones (MNN engine) |
| Phone class | 8 GB RAM or more (TokForge catalog) |
| Download | 993 MB (11 files) |
| Base model | openbmb/MiniCPM-V-4.7-1B |
| License | Apache-2.0 |
TokForge
- Website: https://tokforge.ai
- Discord: https://discord.gg/Acv3CBtfVm
- Google Play: https://play.google.com/store/apps/details?id=dev.tokforge
- iOS TestFlight: https://testflight.apple.com/join/jnufjzRr
Runs on-device in the TokForge app.
Use it in TokForge
- Install TokForge from Google Play.
- On Android, open Models > Recommended > Chat and download MiniCPM-V 4.7 1B (listed from TokForge 1.3.9.5). TokForge lists it on phones with 8 GB RAM or more. On older versions, open Models > Browse, search for
MiniCPM-V-4.7-1B-MNNand download it. - Load it and start a chat. The model runs on the phone; once it is downloaded, chatting needs no connection.
Collections: TokForge Chat Models (MNN).
Files
| File | Size | What it is |
|---|---|---|
config.json |
502 B | MNN runtime settings |
embeddings_bf16.bin |
508 MB | Token embeddings (bf16) |
export_args.json |
1.20 KB | Export settings used for the conversion |
llm.mnn |
2.16 MB | Model graph |
llm.mnn.json |
5.37 MB | Model graph as JSON (reference copy) |
llm.mnn.weight |
470 MB | Model weights |
llm_config.json |
7.78 KB | Model settings and chat template |
tokenizer.mtok |
7.33 MB | Tokenizer |
LICENSE |
11.4 KB | Apache-2.0 license text |
NOTICE |
375 B | Notice crediting OpenBMB and stating the conversion |
SHA256SUMS |
791 B | sha256 of every file |
config.json names tokenizer.mtok (MNN's binary tokenizer). llm.mnn.json is the readable form of the graph.
SHA256SUMS lists the sha256 of every file.
Text only
This bundle carries the language model only. The MiniCPM-V vision encoder is not included, because TokForge's engine has no adapter for it yet. In TokForge, photos are described by the app's built-in image helper instead, and the model answers from that description.
About the model
MiniCPM-V 4.7 1B by OpenBMB, in the small 1B size. Its language model follows the Qwen3.5 design: 24 layers (18 linear attention, 6 full attention), hidden size 1024. This repo holds that language model converted to MNN for phones: 4-bit weights and BF16 embeddings, 0.99 GB of model files. The graph uses the layout that the MNN engine in TokForge 1.3.8.1 and 1.3.9.4 loads.
How it was made
- Source:
openbmb/MiniCPM-V-4.7-1Bat commit379f120c2ff3588e6e06559ec6b9ec15cc05114e. - Converted with MNN's
llmexport(TokForge's pinned build):--quant_bit 4 --quant_block 64 --lm_quant_bit 4 --lm_quant_block 64 --hqq. The exporter change mapped the minicpmv4_7 text backbone onto the existing Qwen3.5 text path; the vision tower and its merger are not exported. - Linear layers, the output head included: 4-bit HQQ, block size 64. Embeddings: BF16, in
embeddings_bf16.bin. - Graph: the plain LinearAttention layout (no
transformer_c4regions, no folded gate).llm_config.jsondeclaresmodel_type: qwen3_5, the language model's own type. - Chat template: OpenBMB's own template, byte for byte (thinking is off unless the request turns it on).
Accuracy on the host
Measured on an x86 build of the MNN engine source that TokForge ships (CPU, with this bundle's own settings), against the original model in float32. Three long real-text prompts (453, 912 and 1,360 tokens) in the model's chat template, scored on the first reply token and 16 more teacher-forced steps each: 51 positions in total.
| measure | result |
|---|---|
| first reply token matches the original | 3 of 3 |
| top-1 token matches the original | 48 of 51 |
| KL divergence to the original, median / mean | 0.025 / 0.120 |
The conversion itself is exact: with 16-bit weights the same graph matches the original at every position, so the small difference that remains comes from the 4-bit weights.
License
Apache-2.0, as the original model. LICENSE holds the license text and NOTICE credits OpenBMB.
Credits
- MiniCPM-V 4.7 1B by OpenBMB (Hugging Face). Used under the Apache-2.0 license.
- Its language backbone builds on Qwen3.5 and Qwen3.6 by the Qwen team (Alibaba Cloud), as OpenBMB credits them (Qwen3.5, Qwen3.6). Apache-2.0.
- Converted with llmexport from MNN by Alibaba (Apache-2.0).
- Downloads last month
- 58
Model tree for darkmaniac7/MiniCPM-V-4.7-1B-MNN
Base model
openbmb/MiniCPM-V-4.7-1B