🤖 Agents-A1-4B-kimi-Preview-GGUF

GGUF Quantized — 4B Coding Agent Model · Kimi K3 Traces · Tool-Calling · Vision

GGUF quantized version of hotdogs/Agents-A1-4B-kimi-Preview — fine-tuned on Kimi K3 coding and debugging traces. Optimized for llama.cpp inference with coding agent capabilities and vision support via mmproj.

---

✨ Key Features

Capability	Description
💻 Coding Agent	Trained on real Kimi K3 coding traces — planning, debugging, building
🖼️ Vision Understanding	Image-text-to-text with mmproj
🧠 Step-by-step Reasoning	Autonomous agent-style reasoning
🔧 Tool Calling	llama.cpp --tools all support
💬 Multi-turn	61% multi-turn conversations
🌏 Thai + English	Native bilingual support
🐍 Multi-language	Python, C, C++, Go, Java, Rust, Bash, and more
⚡ Fast Inference	IQ4_NL fits in ~3 GB VRAM

---

📦 Downloads

File	Size	Description
Agents-A1-4B-kimi-Preview-IQ4_NL.gguf	2.61 GB	Recommended — best quality/speed balance for 8GB VRAM
Agents-A1-4B-kimi-Preview-Q8_0_imatrix.gguf	4.48 GB	Q8_0 + imatrix — almost lossless
Agents-A1-4B-kimi-Preview.gguf	8.42 GB	Full BF16 precision
Agents-A1-4B-mmproj.gguf	672 MB	Vision projector for image understanding

🎯 IQ4_NL is recommended for 8GB VRAM users — fits comfortably even at 128K context with flash-attention.

---

🚀 Usage

Docker (Recommended)

sudo docker run --rm -p 8080:8080 \
  -v /root/models/:/models \
  --gpus all \
  --ulimit memlock=-1:-1 \
  --env CUDA_VISIBLE_DEVICES=0 \
  ghcr.io/ggml-org/llama.cpp:full-cuda --server \
  -m /models/Agents-A1-4B-kimi-Preview-IQ4_NL.gguf \
  --mmproj /models/Agents-A1-4B-mmproj.gguf \
  --host 0.0.0.0 --port 8080 \
  --n-gpu-layers 999 \
  --ctx-size 131072 \
  --batch-size 4096 \
  --ubatch-size 256 \
  --cache-type-k f16 \
  --cache-type-v f16 \
  --flash-attn on \
  --cont-batching \
  --mlock \
  --temp 0.95 \
  --top-k 40 \
  --top-p 0.9 \
  --min-p 0.0 \
  -n -1 \
  --no-mmap \
  --parallel 1 --tools all \
  --dry-multiplier 0.05 \
  --jinja --dry-sequence-breaker none \
  --repeat-penalty 1.1

Parameter Explanation

Parameter	Purpose
--mmproj	Vision projector for image understanding
--ctx-size 131072	128K context window
--flash-attn on	Flash attention for speed
--cache-type-k/v f16	BF16 KV cache for quality
--cont-batching	Continuous batching for multi-turn
--tools all	Enable tool/function calling
--jinja	Use Jinja2 chat template
--mlock	Lock memory for performance

llama.cpp (Direct)

# Quick text-only test
./llama-cli -m Agents-A1-4B-kimi-Preview-IQ4_NL.gguf \
  -p "Write a Python function to sort a list" -n 256 --temp 0.6 -ngl 999

# Vision inference
./llama-cli -m Agents-A1-4B-kimi-Preview-IQ4_NL.gguf \
  --mmproj Agents-A1-4B-mmproj.gguf \
  --image photo.jpg \
  -p "What is in this image?" -n 256 --temp 0.6 -ngl 999

---

🧬 Model Information

This is a GGUF quantized version of hotdogs/Agents-A1-4B-kimi-Preview, which is a fine-tune of InternScience/Agents-A1-4B on coding agent traces.

Parameter	Value
Base Model	hotdogs/Agents-A1-4B-kimi-Preview
Parameters	~4.29B
Architecture	Qwen3.5 hybrid (Linear + Full attention)
Vision	✅ Via mmproj
Context	Up to 128K tokens
Format	ChatML (Jinja2 template)
Fine-tuning	Kimi K3 coding traces (3,389 samples, 3 epochs, scale=0.4)

---

🙏 Acknowledgements / ขอบคุณ

• InternScience — For the Agents-A1-4B base model and mmproj vision projector 🙏
• mmproj source — Extracted from InternScience/Agents-A1-4B-Q4_K_M-GGUF
• greghavens — For the Kimi K3 coding traces dataset
• Qwen Team (Alibaba) — For the Qwen3.5 architecture
• Unsloth AI — For training optimizations
• All dataset contributors and the open-source AI community ❤️

---

💖 Support / โปรดสนับสนุน

If you find this model useful, please consider supporting my work!
หากคุณคิดว่าโมเดลนี้มีประโยชน์ กรุณาสนับสนุนผลงานของฉันด้วยนะคะ! 🙏

₿ Bitcoin — BTC:

bc1qf27cyk3vmugcdyv9xdtuv5jwz37863crpj5c9v

Thank you for your support! 🙏✨
ขอบคุณมากๆ สำหรับการสนับสนุนค่า! 💖🤗

---

Built with ❤️ by UKA — 18-year-old coder & cybersecurity expert