README.md

---
license: other
license_name: blink-research
license_link: LICENSE.md
base_model: Qwen/Qwen3.5-4B
library_name: transformers
pipeline_tag: text-classification
inference: false
tags:
  - decision-model
  - typed-decisions
  - one-pass
  - option-probabilities
language:
  - en
---

# blink-4b

![blink-4b one-pass readout diagram: a state and typed choice, noul and score questions are rendered as evidence, criterion and lettered options; questions are batched and each batch is one forward pass, giving next-token logits at the answer position, and an FP32 softmax over the offered letters gives one probability per option. No text is generated.](assets/blink-4b-readout.png)

Small, fast decisions for routing and checks at volume. Send text or JSON `state` with `choice`, `noul` (yes/no), or `score` questions. Get a probability for every offered answer, not generated text. Each batch takes one forward pass; large requests can use several batches.

Use `choice` to route a request, `noul` for a yes/no check, or `score` for an ordered rating. The same call can ask several questions about a single state.

**Try it:** [Space](https://huggingface.co/spaces/thegovind/blink) · [Screen click](https://huggingface.co/spaces/thegovind/blink?tab=use-cases&case=screen) · [Computer use](https://thegovind.github.io/blink/computer-use/) · [API](https://thegovind.github.io/blink/api/) · [Docs](https://thegovind.github.io/blink/) · [GitHub](https://github.com/thegovind/blink) · [blink-mimo-9b](https://huggingface.co/thegovind/blink-mimo-9b) · [blink-27b](https://huggingface.co/thegovind/blink-27b)

## At a glance

| Attribute | Detail |
|---|---|
| Base model | [Qwen/Qwen3.5-4B](https://huggingface.co/Qwen/Qwen3.5-4B), text weights only |
| Weights size | 8.4 GB (bf16, 4,205,751,296 parameters) |
| Revision | v1.4 (code revision; weights identical to v1.0) |
| License | Weights: non-commercial research and evaluation only ([LICENSE.md](LICENSE.md)); code: Apache-2.0. Base-model notice: Apache-2.0 (`LICENSE-Qwen`). |

## Quickstart

```python
# pip install "torch==2.13.0" "transformers==5.17.0" "flash-linear-attention==0.5.2" "accelerate>=1.1.0" safetensors huggingface_hub
import os, sys
from huggingface_hub import hf_hub_download

os.environ["BLINK_MODEL"] = "thegovind/blink-4b"
os.environ["BLINK_REVISION"] = "v1.4"
sys.path.insert(0, os.path.dirname(hf_hub_download("thegovind/blink-4b", "blink.py", revision="v1.4")))
import blink

out = blink.decide(
    "Order #4411 arrived with a cracked screen. I want my money back, not another one.",
    {
        "intent": {
            "type": "choice",
            "instructions": "What does the customer want?",
            "criteria": {"refund": "Money back", "replacement": "A new unit", "info": "Information only"},
        },
        "urgent": {"type": "noul", "instructions": "Does this need a reply today?"},
        "anger": {"type": "score", "instructions": "How upset is the customer?", "criteria": ["calm", "annoyed", "angry"]},
    },
)
print(out["answers"]["intent"]["probabilities"])
```

## Run it as a server

`serve.py` serves `POST /v1/systemone`, `GET /v1/models`, and `GET /healthz`. Point TypeSafe's server-side Python or JavaScript SDKs at it with `TYPESAFE_BASE_URL`; text decisions use the same request and response fields as hosted Jev. Requests run one at a time by default; `--batch-window-ms 5` enables cross-request batching. The [API reference](https://thegovind.github.io/blink/api/) covers limits and errors.

```sh
pip install "torch==2.13.0" "transformers==5.17.0" "flash-linear-attention==0.5.2" "accelerate>=1.1.0" safetensors huggingface_hub
hf download thegovind/blink-4b --revision v1.4 --local-dir blink-4b
python blink-4b/serve.py --model ./blink-4b --port 8000
# TypeSafe SDKs: export TYPESAFE_BASE_URL=http://127.0.0.1:8000 TYPESAFE_API_KEY=any
```

Or use Docker from the downloaded folder:

```sh
cd blink-4b
docker build -t blink-4b . && docker run --rm --gpus all -p 127.0.0.1:8000:8000 blink-4b
```

## Higher throughput (opt-in)

`serve_vllm.py` (added in `v1.4`) is an opt-in, text-only server with higher throughput. `serve.py` stays the default.

Earlier paired loopback measurements (one self-hosted replica, 231 public items): c1 p50/p95 was 52/127 ms on vLLM vs 64/175 ms on `serve.py` (one request at a time); c16 was 34.7 vs 12.2 completed questions/s (16 concurrent requests).

Earlier peak memory was 68.0 GiB including cache and cold start 180 s; neither was retimed, nor were the other quality sets or c16 rerun. A release build re-passed the 231-item JevBench c1 quality check (80/111 hard, 199/231 total, hard ECE 0.069). Later changes touched only request checks, error handling and startup cleanup, not scoring.

Tested versions:

```sh
python -m pip install "torch==2.13.0" "transformers==5.17.0" \
  "vllm==0.30.0" "compressed-tensors==0.17.0" \
  "accelerate>=1.1.0" safetensors huggingface_hub
python -m pip install "flash-linear-attention==0.5.2"
```

```sh
hf download thegovind/blink-4b --revision v1.4 --local-dir blink-4b
cd blink-4b
python serve_vllm.py --model . --port 8000 --quantization auto --max-concurrency 32 --max-num-seqs 32 --max-num-batched-tokens 8192 --gpu-memory-utilization 0.85
```

[Setup, quality limits and caveats](https://huggingface.co/thegovind/blink-4b/blob/v1.4/VLLM.md). No official JevBench score for blink has been published.

## Screenshots (opt-in, self-hosted)

Image input is off by default. Start `serve.py` with `--vision-tower Qwen/Qwen3.5-4B@851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a` to attach the matching tower. This is a self-hosted blink extension. TypeSafe's hosted Jev is text-only.

With `--vision-tower`, self-hosted blink borrows the pinned Qwen base model's vision encoder while its checkpoint stays text-only; [blink-mimo-9b](https://huggingface.co/thegovind/blink-mimo-9b) uses its own encoder and runs [Screen click](https://huggingface.co/spaces/thegovind/blink?tab=use-cases&case=screen).

Put a `data:image/png;base64,...` URI (JPEG and WebP data URIs work too) inside a `state` string, or pass data URIs in a top-level `images` list. Image URLs are never fetched. If loading only `blink.py` via `hf_hub_download`, also download `graft_keys.py` from the same repo and revision beside it.

See [Computer use](https://thegovind.github.io/blink/computer-use/) to self-host this model with images.

## Results

| Local development readout | Result |
|---|---:|
| JevBench public-items proxy | 76.5; 80/111 hard; hard ECE 0.067 |
| Decision Index 0.2 balanced skill | 37.85 |

No official JevBench score for blink has been published. The JevBench numbers are public-item development proxies, not an official score, rank, or parity claim. Decision Index is a descriptive local run of the official kit, not a leaderboard submission; training overlap affects its scores.

<details><summary>Model details: architecture, training, data</summary>

### Architecture and readout

![blink-4b network diagram: 32 decoder layers repeating 3 Gated DeltaNet layers then one full-attention layer (24 and 8 in total, hidden 2560), with full attention at 0-based layers 3, 7, … 31 as in the tensor names; LoRA rank 16 on every attention, Gated DeltaNet and MLP projection (32.5M parameters, merged after training); token embeddings, norms and lm_head frozen, with lm_head tied to the token embeddings, one matrix; the vision encoder and multi-token-prediction head removed; and the answer read from the offered option-letter rows of lm_head.](assets/blink-4b-network.png)

Qwen3.5-4B text backbone: 32 decoder layers (24 Gated DeltaNet, 8 full-attention), hidden 2560, and 4,205,751,296 shipped parameters. The training updated 32.5M LoRA parameters at rank 16, alpha 32: `q_proj`, `k_proj`, `v_proj`, `o_proj`; `in_proj_qkv`, `in_proj_z`, `in_proj_a`, `in_proj_b`, `out_proj`; and `gate_proj`, `up_proj`, `down_proj`. The vision encoder and MTP head were removed: 0 vision tensors, 0 MTP tensors.

Readout softmaxes FP32 next-token logits over the offered option labels. These are option-conditional probabilities, not certified chances of being right.

### Training

![blink-4b post-training diagram: T3 (23,156 rows) and T4 (42,360 rows) broken down by data category feed supervised fine-tuning with cross-entropy over the offered option letters; both runs start from the base, and T3, T4 step 300 and T4 final are merged and averaged into release v1.0. Decision Index 0.1 full suite 52.12.](assets/blink-4b-post-training.png)

Supervised fine-tuning on target distributions; no RL or preference optimization. T3 used lr 3e-5 for 96 steps; T4 used lr 4e-5 for 472 steps. The released weights average T3, T4 step 300, and T4 final.

| Stage | Question rows | Mix |
|---|---:|---|
| T3 | 23,156 | 11,352 decision worlds · 3,741 teacher-written rows · 2,579 exact-probability worlds · 2,221 public-source · 1,763 base-model anchors · 1,500 program-generated reasoning |
| T4 | 42,360 | 12,000 decision worlds · 7,860 teacher-written rows · 7,000 judge-style · 6,220 exact-probability worlds · 3,500 base-model anchors · 3,000 program-generated reasoning · 2,780 public-source |

### Data sources and licences

| Source | Licence |
|---|---|
| MMLU auxiliary train, MMLU-Pro, CommonsenseQA, GSM8K | MIT |
| AQuA-RAT, Amazon ESCI | Apache-2.0 |
| MedMCQA | Apache-2.0 (dataset card) |
| SuperGPQA | ODC-BY |
| WANLI, ContractNLI, BANKING77, GPQA | CC BY 4.0 |
| ARC | CC BY-SA 4.0 |
| BoolQ, Dolly-15k | CC BY-SA 3.0 |
| ANLI | CC BY-NC 4.0 |
| SciQ | CC BY-NC 3.0 |
| iSarcasmEval | MIT (upstream repository licence) |
| VAST, Humicroedit, OpenBookQA | None stated by source |
| Code-generated worlds and teacher-written documents (Qwen3.8-27B) | See LICENSE.md |

Source-repository licences do not settle rights in every underlying text.

### Evaluation and limits

The archived Decision Index 0.1 local run scored 52.12; editions are not directly comparable. Public JevBench items were used for development selection, not training. Training included 281 MMLU-Pro test-partition questions, contaminating that Decision Index 0.2 component; semantic and pretraining overlap cannot be ruled out. English-centric, no chatting or explanations; text in `state` can sway an answer, especially on long policies and date arithmetic.

For `blink.py` and default `serve.py`: 255 options per choice, 2–10 score levels, 131,072 input tokens per question, and 512 questions per request; over-limit requests return 422 without truncation. For image placement use `--image-layout first|inline` (`first` is the default); for a renamed folder use `--model-name blink-4b`. Neither flag enables images on its own.

</details>

## License

Code: Apache-2.0. Weights: non-commercial research and evaluation only; see each model card's license.

See [LICENSE.md](LICENSE.md) for weight terms; the Qwen base model is Apache-2.0 (`LICENSE-Qwen`).