Nemotron-Labs-Diffusion-VLM-8B-GGUF
Audited community GGUF conversion for research with AR, 32-token block diffusion, and linear self-speculation through the supplied runner. This is not an official NVIDIA or Ollama release. The non-AR modes require attention and cache control provided by the accompanying runner.
Includes the separately audited Pixtral vision encoder and projector. Text generation also works without loading it.
Non-commercial research/evaluation only under NVIDIA NSCLv1. This also applies to the text-only derivative. See the complete license and the unchanged source model card.
Files
| File | Size (decimal GB) | Storage |
|---|---|---|
| VLM8B-F16.gguf | 16.988 | F16 |
| mmproj-F16.gguf | 0.857 | F16 vision |
Excluded from this release: Q8_0, Q5_K_M, Q4_K_M. Only sizes passing the declared release gates are included.
Choose one language GGUF. The full VLM additionally needs mmproj-F16.gguf
for images. Tokenizer files and the chat template are included. F16 baselines
are preserved; no importance matrix or further training was used.
Runtime support
| Runtime | AR text | Block diffusion | Self-speculation | Images |
|---|---|---|---|---|
| Supplied pinned runner | Tested | Tested | Tested, sequential verifier | Tested |
| Stock Ollama 0.33.3 | F16 import/raw/chat/multi-turn tested | Not implemented | Not implemented | Not claimed |
Ollama runs only the AR path. It cannot acquire diffusion or self-speculation from a Modelfile. Vision input is validated in the supplied runner. Stock Ollama vision is not a validated compatibility claim.
Run
Build the small bridge using runtime/README.md, with pinned
llama.cpp b10760. From the downloaded repository directory:
python runtime/generate.py --model VLM8B-F16.gguf --tokenizer . \
--mode diffusion --prompt 'Explain how rain forms.' --max-new-tokens 128
python runtime/generate.py --model VLM8B-F16.gguf --tokenizer . \
--mode self-speculation --prompt 'What is 2 + 2?' --max-new-tokens 32
ollama create nemotron-vlm8b-ar -f Modelfile.F16
curl http://localhost:11434/api/chat -d '{"model":"nemotron-vlm8b-ar","messages":[{"role":"user","content":"What is 2 + 2?"}],"think":false,"stream":false}'
The runner and tested Ollama server use GGML_CUDA_CUBLAS_COMPUTE_TYPE=f32.
Set it in the server environment for Ollama; it is not a Modelfile parameter.
The runner sets it by default. F16 weights/KV remain F16. Read the runtime
documentation before overriding this measured precision requirement.
Ollama requests must explicitly set think: false for the tested non-thinking
behavior; the default thinking mode is not covered by this compatibility result.
Image example, after building the bridge:
python runtime/generate.py --model VLM8B-F16.gguf --tokenizer . \
--projector mmproj-F16.gguf --image example_images/red.png \
--mode diffusion --prompt 'What is the main color?' --max-new-tokens 32
Validation and limits
All original language tensors match after documented Q/K permutation and dtype conversion. The full VLM's vision tensors are audited separately. Tokenizer and positional metadata checks pass. FP16 language checks have a maximum relative L2 logit error of 0.000957 against the pinned NVIDIA FP16 eager reference under the recorded test environment.
FP16 generation across 8 selected text/image fixtures matches the NVIDIA reference for all three modes: true. This is a limited regression result, not a general cross-backend equality claim. Detailed outputs and all numerical checks are in reports.
The 973-token, 12-probe causal regression suite is author-created, with prose, code, mathematics, and multiple languages. It is not a standardized benchmark or a downstream accuracy estimate.
| Storage | Mean KL vs F16 | NLL increase | Top-1 agreement |
|---|---|---|---|
| F16 | 0.000000 | +0.000000 | 100.0000% |
Quantized outputs may differ. Sequential self-speculation is checked against each quant's own AR output, not required to reproduce F16 text. The optional batched verifier has separate equality measurements in each validation report.
Important boundaries:
- Greedy decoding only. Self-speculation defaults to one-token AR verification, retaining exactly the verified prefix. No throughput improvement is claimed.
--verification batchedis an experiment; changing batch shape may change close floating-point token decisions.- This linear verifier does not reproduce the upstream VLM quadratic SBD lattice, draft LoRA, or distribution-preserving stochastic speculation.
- Positional checks near 16384/32768 use synthetic consecutive short prefixes, not a full long-context benchmark. The example context is 4096.
- Image tests use simple synthetic fixtures, not an OCR/VQA benchmark.
- Thinking, tools, audio, video, streaming, and other GPUs/CPUs are not covered by these release claims. The Ollama template targets plain non-thinking chat.
Provenance
- Upstream:
nvidia/Nemotron-Labs-Diffusion-VLM-8B@adca93d16471c1e07d594ae444d23e1876f6b365. - llama.cpp:
0f3a71be15af836d277c9f918adfafb45732677e. - Conversion/runtime source commit:
e962ef052056fc51ba704dd5b65b6136236d048b. - Hardware: NVIDIA A100-SXM4-40GB; full dependencies in reports/environment.json.
- Full file hashes: SHA256SUMS, manifest.json.
Model weights and upstream tokenizer assets retain NVIDIA's terms. The original IntellAgents runner/conversion additions use MIT; that does not relicense the model. Conversion changes tensor layout, storage type, metadata, and optional quantization; it performs no training.