rockypod/neotoi-coder

Neotoi Coder v2.0

A Rust/Dioxus 0.7 specialist fine-tuned from Qwen3-Coder-14B using RAFT (Retrieval-Augmented Fine-Tuning). Optimized for production-quality Dioxus 0.7 components with Tailwind v4, WCAG 2.2 AAA accessibility, GlobalSignal state management, i18n, dark mode, and static content navigation.

What's New in v2.0

Significant improvements over v1.0 across every tier:

• New Tier 8 — GlobalSignal & i18n: Correct .write() vs .set() semantics, pre-rsx! i18n bindings, dark mode via document::eval, sticky nav with use_hook
• New Tier 9 — Static Content Navigator: use_memo filtering, tag-based routing, deterministic LLM intent mapping over static content
• New Tier 10 — Dioxus 0.7.4 APIs: WritableResultExt, use_context panic behavior, WebSocket Stream+Sink, consume_context vs use_context
• T6 Hard Reasoning: Went from 2/5 passes to 10/10 — full perfect score
• T4 WCAG/ARIA: Perfect 14/14 — tooltip always in DOM, listbox/option nesting, aria_labelledby on all role containers
• T5 use_resource: Perfect 8/8 — three-arm match, no .ok() wrapper, signal read inside closure
• MLX format included for Apple Silicon (Ollama 0.19+, mlx_lm)

Exam Results

v2.0 — 100 Question Weighted Exam

Tier	Questions	Weight	Score	Max	Status
T1 Fundamentals	Q1–12	1.0	11/12	12	✅
T2 RSX Syntax	Q13–24	1.0	10/12	12	✅
T3 Signal Hygiene	Q25–36	1.0	12/12	12	✅ Perfect
T4 WCAG/ARIA	Q37–50	1.5	14/14	21	✅ Perfect
T5 use_resource	Q51–58	1.5	8/8	12	✅ Perfect
T6 Hard Reasoning	Q59–68	2.0	10/10	20	✅ Perfect
T7 Primitives+CSS	Q69–80	1.5	11/12	18	✅
T8 GlobalSignal/i18n	Q81–88	1.5	8/8	12	✅ Perfect
T9 Static Navigator	Q89–94	1.5	6/6	9	✅ Perfect
T10 Dioxus 0.7.4	Q95–100	2.0	6/6	12	✅ Perfect
Overall	Q1–100		135.5/140	140	✅ 96.8%

Version History

Version	Score	Exam	Status
v1.0	51/60 (85%)	60Q standard	Published
v2.0	135.5/140 (96.8%)	100Q weighted	Published

Model Details

• Base model: Qwen3-Coder-14B
• Method: RAFT (Retrieval-Augmented Fine-Tuning)
• Dataset: 4,185 curated Dioxus 0.7 examples (3 training runs)
• Scope: Rust + Dioxus 0.7 + Tailwind v4 + WCAG 2.2 AAA
• Quantization: Q4_K_M (8.4 GB) and MLX 4-bit (7.8 GB)
• Author: Kevin Miller, Jr.

Read the Full Story

Detailed writeup on the training methodology, exam results, and infrastructure:

Neotoi Coder v2.0 — RockyPod.com

---

Files

File	Format	Size	Use case
neotoi-coder-v2-q4_k_m.gguf	GGUF Q4_K_M	8.4 GB	LM Studio, llama.cpp, Ollama Linux
mlx/	MLX 4-bit	7.8 GB	Ollama 0.19+ Apple Silicon, mlx_lm
neotoi-coder-v1-q4_k_m_final.gguf	GGUF Q4_K_M	8.4 GB	v1.0 legacy

Enabling Thinking Mode

LM Studio

In the chat interface go to the prompt template settings and configure:

Field	Value
Before System	<|im_start|>system
After System	<|im_end|>
Before User	<|im_start|>user
After User	<|im_end|>
Before Assistant	<|im_start|>assistant\n<think>
After Assistant	<|im_end|>

Ollama (Linux / GGUF)

FROM neotoi-coder-v2-q4_k_m.gguf
PARAMETER temperature 0.2
PARAMETER num_predict 4096
PARAMETER repeat_penalty 1.15
PARAMETER stop "<|im_end|>"
TEMPLATE """<|im_start|>system
{{ .System }}<|im_end|>
<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
<think>
"""
SYSTEM You are Neotoi, an expert Rust and Dioxus 0.7 developer. Always think step-by-step before answering.

mlx_lm server (Apple Silicon / Ollama 0.19+)

mlx_lm server \
  --model path/to/mlx/ \
  --host 127.0.0.1 \
  --port 8080

Then point any OpenAI-compatible client at http://localhost:8080.

llama.cpp

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

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 (simulated only — v2.1 target)

License

Neotoi Coder Community License v1.0 — see LICENSE file. Commercial use of model outputs permitted. Weight redistribution prohibited. Mental health deployment requires written permission.

Credits

Built with:

• Unsloth — 2x faster fine-tuning
• TRL — SFTTrainer
• Qwen3-Coder-14B — base model
• MLX — Apple Silicon inference
• Claude Code — dataset pipeline and training infrastructure
• Dioxus — the framework this model specializes in