GGUF quantizations of deepseek-ai/DeepSeek-R1-0528-Qwen3-8B, fine-tuned on the v3 synthetic patent-reasoning corpus and verified end-to-end on the NVIDIA DGX Spark (GB10, 128 GB unified memory).

What this model does

Offline patent-prosecution reasoning on Spark-class hardware

Patent prosecution work — claim construction, MPEP-grounded office-action responses, Markush analysis, doctrine-of-equivalents reasoning — happens inside firms that can't ship privileged client text to a hosted frontier API. This release distills DeepSeek-R1's chain-of-thought reasoning onto a 5,000-row synthetic patent-reasoning corpus so a single Spark-class box can run the workflow offline, with full IRAC-shaped reasoning chains.

Use cases:

Claim construction (Markush groups, doctrine of equivalents)
MPEP-grounded office-action argument drafting
Prior-art relevance + non-obviousness reasoning chains
Patent-licensing scenario analysis (most-favored-licensee, FTO)

Who this is for: Patent attorneys, prosecution-team engineers, and IP-strategy teams running privileged workflows offline on Spark-class hardware (GB10, 128 GB unified memory) or comparable edge devices.

Notebooks

Two runnable notebooks ship with this model — open either on a free cloud GPU:

Notebook	What it does	Open
Builder	Reproduce this model's build and DGX Spark benchmarks end-to-end with `fieldkit`.	Open In Colab Open in Kaggle
User	Load the published model and call it from your own app in a few lines.	Open In Colab Open in Kaggle

Spark-tested

Every Orionfold quant ships with a measurement triple on the NVIDIA DGX Spark (GB10, 128 GB unified memory): perplexity, sustained tok/s, and thermal envelope. The numbers below are the actual run, not a wishlist.

Variant	Size	Perplexity (wikitext-2)	tok/s on Spark
Q4_K_M	4.68 GB	10.242	39.6
Q5_K_M	5.45 GB	10.044	35.0
Q6_K	6.26 GB	9.962	30.7
Q8_0	8.11 GB	9.929	26.5

Variants

Variant	Recommended use
Q4_K_M	Smallest footprint with workable quality — fits on a 6 GB GPU; pick this for tightest RAM budgets.
Q5_K_M	Recommended — best quality/throughput balance on Spark (32–35 tok/s, perplexity within 2% of Q8_0).
Q6_K	Near-lossless quality with modest size bump; pick if you have headroom.
Q8_0	Effectively lossless; reach for this when reasoning-chain fidelity matters more than throughput.

Choosing this lane

llama.cpp-quantized variants of the NeMo Framework-trained checkpoint — the bakeoff-winning lane. Sustained 35.0 tok/s on Spark at Q5_K_M, perplexity 10.04 (-11.5% vs the Unsloth baseline on the same recipe), mean reasoning chain 1,320 tokens (+44%). Default pick for Spark-class offline patent reasoning — and the only published lane: the Unsloth baseline was measured for the bakeoff (see the article) but is not published as a downloadable artifact.

How to run

Pull a variant:

bash

huggingface-cli download Orionfold/patent-strategist-v3-nemo-GGUF model-Q5_K_M.gguf \
  --local-dir ./models/patent-strategist-v3-nemo

Serve it via llama-server (OpenAI-compatible API):

bash

llama-server -m ./models/patent-strategist-v3-nemo/model-Q5_K_M.gguf \
  -c 4096 -ngl 99 -t 8 \
  --host 0.0.0.0 --port 8080

Or run in-process via llama-cpp-python:

python

from llama_cpp import Llama
llm = Llama(
    model_path="./models/patent-strategist-v3-nemo/model-Q5_K_M.gguf",
    n_ctx=4096, n_gpu_layers=99,
)
out = llm.create_chat_completion(
    messages=[{"role": "user", "content": "A licensee under a non-exclusive patent license discovers the licensor has signed a more favorable royalty rate with a later licensee. The agreement contains a most-favored-licensee clause requiring rate parity. Walk through the legal and commercial steps the original licensee should take to enforce parity, including notice requirements and remedies."}],
    temperature=0.0,
)
print(out["choices"][0]["message"]["content"])

LM Studio and Ollama (via a Modelfile) load the GGUF directly with no additional setup.

Methods

Full methodology and Spark-side measurement protocol: Two paths to the same chain — Unsloth vs NeMo Framework on Spark.

Known drift

Bounded limitations observed during Spark-side measurement. Each item below names the artifact and the scope of the drift; the balance of the bench measures clean — see Methods for the full breakdown.

"metes-and-times" terminology — Two known terminology drifts inherited from the v3 synthetic corpus; balance of probe answers (~99%) cite real MPEP sections. Correct legal term in claim construction is metes and bounds.
Fabricated MPEP §2163.05(s) citation — Same scope — corpus-generator artifact, not a model-wide hallucination pattern. Real §2163.05 has subsections (a)–(f) on written-description support; subsection (s) does not exist.

Other Orionfold variants

Sibling repos from the same release:

Variant	Lane	Format
`Orionfold/patent-strategist-v3-nemo`	NeMo Framework	BF16 (transformers)
`Orionfold/patent-strategist-v3-nemo-GGUF`	NeMo Framework	GGUF (llama.cpp)

Published by Orionfold LLC · orionfold.com · Methods documented at ainative.business/field-notes.