README.md

metadata

license: other
license_name: neotoi-coder-community-license
language:
  - en
  - vi
base_model: Qwen/Qwen3-Coder-14B
tags:
  - rust
  - dioxus
  - dioxus-0.7
  - rsx
  - fine-tuned
  - raft
  - code
  - mlx
  - gguf
  - wcag
  - accessibility
  - tailwind
  - unsloth
  - qwen3
  - local-llm
  - continue-dev
pipeline_tag: text-generation
model-index:
  - name: neotoi-coder-v2
    results:
      - task:
          type: text-generation
        metrics:
          - type: custom
            name: Dioxus 0.7 Weighted Exam (100Q)
            value: 96.8

Neotoi Coder v2.0

A Rust/Dioxus 0.7 specialist LLM — 96.8% on a 100-question weighted exam. Built with RAFT on a homelab RTX 3090 Ti. No cloud GPUs.

Read the whole story on RockyPod.com → Companion GitHub repo — benchmarks and integration guides

Direct Download

No account or approval required.

File	Size	Format
neotoi-coder-v2.0-q4_k_m.gguf	8.4GB	GGUF Q4_K_M
mlx/ weights	7.8GB	MLX 4-bit

Quick Start

Apple Silicon (mlx_lm)

pip install mlx-lm
mlx_lm server --model /path/to/neotoi-v2.0-mlx --port 8081

Linux (Ollama)

ollama create neotoi-coder-v2 -f Modelfile
ollama run neotoi-coder-v2

LM Studio

Download neotoi-coder-v2.0-q4_k_m.gguf above. See the LM Studio setup guide.

Continue.dev

See the Continue.dev config.

Zed Editor

See the Zed setup guide.

Exam Results

Tier	Score	Max	Status
T1 Fundamentals	11/12	12	✅
T2 RSX Syntax	10/12	12	✅
T3 Signal Hygiene	12/12	12	✅ Perfect
T4 WCAG/ARIA	14/14 × 1.5	21	✅ Perfect
T5 use_resource	8/8 × 1.5	12	✅ Perfect
T6 Hard Reasoning	10/10 × 2.0	20	✅ Perfect
T7 Primitives+CSS	11/12 × 1.5	18	✅
T8 GlobalSignal/i18n	8/8 × 1.5	12	✅ Perfect
T9 Static Navigator	6/6 × 1.5	9	✅ Perfect
T10 Dioxus 0.7.4	6/6 × 2.0	12	✅ Perfect
Total	135.5/140	140	96.8%

What It Knows

• Dioxus 0.7 RSX brace syntax — never function-call style
• use_signal, use_resource with correct three-arm match
• r#for on label elements only, never inputs
• GlobalSignal — .write() not .set() for statics
• WCAG 2.2 AAA: tooltip always in DOM, listbox/option nesting, aria_labelledby on all role containers
• dioxus-primitives — no manual ARIA on managed components
• styles!() macro for CSS modules
• Tailwind v4 utility classes and semantic tokens
• EN/VI i18n via pre-rsx! let bindings
• Dark mode via document::eval + CSS custom properties
• Static content navigation with use_memo filtering
• use_context panics without provider — never returns None
• WritableResultExt from Dioxus 0.7.4

What It Does Not Know

• Playwright/E2E testing (out of scope)
• Non-Dioxus web frameworks
• WebSocket Stream+Sink real patterns (v2.1 target)

Enabling Thinking Mode

LM Studio

Field	Value
Before System	`<
After System	`<
Before User	`<
After User	`<
Before Assistant	`<

llama.cpp

./llama-cli \
  -m neotoi-coder-v2.0-q4_k_m.gguf \
  -ngl 99 \
  --temp 0.2 \
  -p "<|im_start|>user\nYour question<|im_end|>\n<|im_start|>assistant\n<think>"

Model Details

• Base model: Qwen3-Coder-14B
• Method: RAFT (Retrieval-Augmented Fine-Tuning)
• Dataset: 4,185 curated Dioxus 0.7 examples
• Training: 4 epochs, RTX 3090 Ti, ~4 hours
• Train loss: 0.3727 (from clean Qwen3-14B base)
• Quantization: Q4_K_M (8.4 GB) and MLX 4-bit (7.8 GB)

License

Neotoi Coder Community License v1.0. Commercial use of outputs permitted. Weight redistribution prohibited.

Credits

Built with Unsloth, Qwen3-Coder-14B, MLX, and Claude Code.