Instructions to use tjarvis91/vfai-x-3.5-9b-options with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tjarvis91/vfai-x-3.5-9b-options with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="tjarvis91/vfai-x-3.5-9b-options")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("tjarvis91/vfai-x-3.5-9b-options")
model = AutoModelForMultimodalLM.from_pretrained("tjarvis91/vfai-x-3.5-9b-options")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use tjarvis91/vfai-x-3.5-9b-options with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "tjarvis91/vfai-x-3.5-9b-options"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tjarvis91/vfai-x-3.5-9b-options",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/tjarvis91/vfai-x-3.5-9b-options

SGLang

How to use tjarvis91/vfai-x-3.5-9b-options with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "tjarvis91/vfai-x-3.5-9b-options" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tjarvis91/vfai-x-3.5-9b-options",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "tjarvis91/vfai-x-3.5-9b-options" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "tjarvis91/vfai-x-3.5-9b-options",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use tjarvis91/vfai-x-3.5-9b-options with Docker Model Runner:
```
docker model run hf.co/tjarvis91/vfai-x-3.5-9b-options
```

🟣 Try the Qovaryx Options Decoder — Free beta

This model is part of Qovaryx, the sovereign 11-head AI cluster that grades options trades in under a millisecond on your CPU.

🚀 Download the desktop app (beta): https://qovaryx.jehorizon.com/download.html

📖 Read the research: https://qovaryx.jehorizon.com/research

💬 Community Discord: https://discord.gg/PtuHZDv5ju

🌐 Main site: https://qovaryx.jehorizon.com

Not financial advice. Options trading involves substantial risk. You make the call.

⚠️ Deprecated — superseded by the Qovaryx app

This model is retained for reproducibility but is no longer the recommended way to run trading inference. The signed, watermarked Qovaryx beta ships from our site only — not from Hugging Face — so we can control versions and apply the nuke-switch if a build needs to be retired.

👉 Get the app: https://qovaryx.jehorizon.com/download.html

Download VFAi-X Qovaryx 2.5.3 Hotfix

Auto-update channel: updates/latest.json now points to 3.5v31-2.5.3-qovaryx-public. Existing public installs should pull the Qovaryx 2.5.3 hotfix on startup.

This hotfix fixes Qovaryx sidecar AI status, removes the legacy vLLM/Docker/WSL waiting path when Qovaryx is healthy, repairs the bundled CPU Torch runtime, restores Electron V8 startup snapshots, preserves user-entered Tradier/API keys, prevents balance/positions blinking during transient Tradier refresh failures, repairs stale Qovaryx runtime downloads, and keeps the public build keyless.

Model/runtime source: Qovaryx/qovaryx-options-decoder-full-community.

Installer Mirror

Auto-update pointer: updates/latest.json

This is the same 2.5.3 public hotfix listed above. It ships no API keys or broker credentials. Existing installs pull this update from the Hugging Face public update channel.

🚀 New flagship: Qovaryx Options Decoder — Full Community Runtime. The latest, most capable Qovaryx release is live as a single drop-in package: six functional HGB specialists + eight vaulted torch heads in one runtime. 15-of-15 internal benchmark cells closed at strict bootstrap CI lower bound. Drop-in replacement for FrankenB / V3.7 / Qwen-VPA. Sub-millisecond inference. Offline. No license email required. 👉 Qovaryx/qovaryx-options-decoder-full-community

💬 Join the community. Discord: https://discord.gg/PtuHZDv5ju — builders training their own trading/finance models. Engineering, no signals. Get install help, share work, follow the Qovaryx research devlog. Try the deployed Q-Chat router live via /qchat ask. Ko-fi: https://ko-fi.com/tjarvis91 — every coffee literally buys GPU time for the next training cycle.

VFAi-X 3.5 9B Options -- Legacy Lineage (V3.7 + V6 Sniper)

Legacy lineage repo for VFAi-X. The current shipping desktop app (VFAi-X Qovaryx 2.5.3, version 3.5v31-2.5.3-qovaryx-public) runs the Qovaryx Options Decoder community runtime — sub-millisecond CPU inference, no GPU required. The Franken-B vision-language flagship at tjarvis91/vfaix-vpa-options-trader is the previous runtime engine and is now part of the historical lineage alongside V3.7 and V6 Sniper. This repo is kept live for (1) the installer auto-update channel and existing installed users, and (2) two model variants that remain useful as research / ensemble components for anyone studying the lineage.

Live branches on this repo:

Branch	Model	Role
`main` / `v3.7`	VFAi-X V3.7	Prior flagship -- best small-N corpus coverage, best fusion text-only, brutal action_pass 100%
`vfai-x-sniper-options`	VFAi-X V6 Sniper	High-conviction abstention specialist -- highest profit-factor in the family, ~2% raw take-rate

The Franken-B installer pulls the new flagship by default. V3.7 and V6 Sniper are opt-in for research, ensembling, or operators who want them as a confirmation filter alongside Franken-B.

App Showcase

See what the desktop app looks like before you install. (AI engine off in screenshots; connects automatically on first launch.)

Dashboard -- account balance, buying power, confidence panels, system status with GPU readout.	Charts -- per-ticker chart with model-emitted Breakout / SH / SL annotations. Penny Universe sidebar.
Orders -- AI Trading Activity: open / filled / cancelled / partial, ticker filter, time-range tabs.	Options -- full options chain with IV / delta / OI / vol / bid / ask. Portfolio Greeks. Risk Analysis.
Fundamentals -- per-ticker valuation, margins, key highlights, macro + earnings calendars.	Change Log -- release notes, runtime checks, local diagnostics, backend tools.

Historical download block — superseded

Auto-update pointer used by the app: updates/latest.json

Current released version (served by both this legacy repo and the flagship repo): 2.5.3 / 3.5v31-2.5.3-qovaryx-public

What Ships

Local-first Windows desktop app with vLLM acceleration.
Franken-B flagship model path and served-model lock.
Hybrid stocks/options operation with constrained penny mode enabled by default.
Tradier broker bridge for user-supplied paper/live credentials.
Public-safe settings: no bundled broker, Hugging Face, data-provider, or private keys.
Hugging Face auto-update channel for critical hotfixes.

Franken-B Runtime Lock

Position sizing: 20% max position size, 8 max positions.
Options overlay: 15x leverage assumption, -5% hard stop, 0.5% theta drag.
Signal gates: 70 conviction floor, 1.0 relative-volume floor, 0.70 profit-score floor.
Fusion path: vision, fusion, gems, chart analysis, and position-manager vision remain enabled.
Thinking mode remains off for this public runtime profile.

Install Notes

The installer is about 1.1 GB and does not bundle the model weights. On first launch it downloads the current Franken-B model files from Hugging Face, verifies the 6 safetensor shards and processor/tokenizer files, then starts vLLM.

If the model repo is ever private or gated, add a Hugging Face read token in Settings or set HF_TOKEN before first launch. Public builds ship no secrets by design.

Support

Independent training runs cost real GPU time. Support is optional and never gates the app: ko-fi.com/tjarvis91

V3.7 -- Prior Flagship (still useful as opt-in)

V3.7 was the production flagship before Franken-B. It is the model the existing 800+ installed users have been running, and it remains a strong reference for text-heavy reasoning and small-N corpus discipline.

Where V3.7 still wins (vs Franken-B):

Metric	V3.7	Franken-B	Read
Brutal action_pass	100% (33/33)	90.9%	V3.7 wins discipline gate cleanly
App-integrated	99.75%	99.0%	tied for practical purposes
Corpus avg (9 packs)	96.2%	88.2%	V3.7 wins broad small-N corpus coverage
Corpus position_focus	100%	90.0%	V3.7 perfect on position cases
Corpus v4	99.0%	91.0%	V3.7 +8pp
Corpus v2_100	98.0%	89.0%	V3.7 +9pp
Fusion text_only	+78.19%	+64.86%	V3.7 wins text-only fusion

V3.7 is the right answer when the input is text-dense and the task rewards small-N discipline. Franken-B wins when chart fusion matters and when stream-level profit-factor matters.

Load V3.7 directly:


vllm serve tjarvis91/vfai-x-3.5-9b-options --revision v3.7 \

  --served-model-name vfai-qwen35-9b-v37 \

  --max-model-len 4096 --gpu-memory-utilization 0.85 \

  --quantization fp8 --trust-remote-code

V6 Sniper -- Companion Abstention Specialist (the `vfai-x-sniper-options` branch)

V6 is a high-conviction abstention specialist. It refuses to emit a trade on ~98% of opportunities. On the rare cases it does fire, it produces the highest per-trade profit factors in the entire VFAi-X family -- by a wide margin.

Why a 2%-take-rate model has shelf space:

Stream	V6 trades	V6 PF	Franken-B PF	V3.7 PF
2yr	432	5.13	4.51	2.99
penny	466	37.18 🏆	31.77	26.77
180d	193	7.71 🏆	4.95	4.75

V6's penny PF of 37.18 is the highest profit factor we have ever measured in any model in the lineage. It comes with the trade-off you would expect: V6 only takes 0.2% (penny) to 2.0% (180d) of raw stream opportunities. It is not a primary emitter -- it is a confirmation filter that pairs well with Franken-B or V3.7.

Corpus-side V6 wins:

Corpus adversarial: 100% (Franken-B 83.3%, V3.7 90.0%)
Corpus combined: 96.67% (Franken-B 83.3%, V3.7 88.3%)

Suggested deployment role:


primary emitter (Franken-B or V3.7) -> emits BUY/SELL

       -> if V6 Sniper also agrees: high-conviction setup, upsize

       -> if V6 emits NO_TRADE: primary wants to act, V6 sees something it missed -- downsize or skip

       -> if V6 emits opposite direction: models disagree, skip

We are not yet shipping the ensemble inside the desktop app -- V6 is published for researchers and operators who want to wire it themselves.

Load V6 Sniper directly:


vllm serve tjarvis91/vfai-x-3.5-9b-options --revision vfai-x-sniper-options \

  --served-model-name vfai-qwen35-9b-v6-sniper \

  --max-model-len 4096 --gpu-memory-utilization 0.85 \

  --quantization fp8 --trust-remote-code

V6 audit thread on HF Discussions: V6 Sniper -- methodology + per-trade evidence

Honest caveats (applies to all three branches)

All overlay-PnL numbers are simulated backtests against locked answer keys we have audited and found to be selection-biased (286,000 directional rows, zero losers by construction). They are upper bounds and best read as relative model-vs-model comparisons on identical conditions, not as forward returns.
The matched-pair (apples-to-apples) comparisons are the most defensible single number on each branch.
Walk-forward validation on a fresh distribution-realistic hold-out is queued for all three branches.
This is research output, not financial advice. Options trading can lose 100% of premium.

Where to go from here

Most users: install the desktop app (link above). It pulls Franken-B by default and you can forget about the model selection entirely.
Researchers / ensemble builders: load V3.7 or V6 Sniper directly with the commands above.
New flagship + audit-grade methodology + composition-lock writeup: tjarvis91/vfaix-vpa-options-trader
Devlog series (1-7, plus launch audit): Discussion threads on this repo and the flagship repo.
Support the next training run: ko-fi.com/tjarvis91

Apache 2.0. Research artifact. Not financial advice. Past evaluation performance does not guarantee future performance.

-- VFAi-X. Independent behavioral-model systems lab. Local-first. Open-source.

Downloads last month: 1,413

Safetensors

Model size

9B params

Tensor type

BF16

Model tree for tjarvis91/vfai-x-3.5-9b-options

Merges

1 model

🟣 Try the Qovaryx Options Decoder — Free beta

⚠️ Deprecated — superseded by the Qovaryx app