APEX Vision Agentic ⚠ Template Fix

Nex-N2-mini

Agentic Vision MoE — APEX Quantized GGUF (Stock llama.cpp)

⚠️ Temporary Workaround — Not Official

👉 Want the unmodified, recommended version? Go to SC117/Nex-N2-mini-APEX-GGUF (use with Nex's patched llama.cpp)

This is a temporary, unofficial workaround for stock llama.cpp users. The original chat_template.jinja has been replaced with a fixed version so that --reasoning-format works without --chat-template-file.

⚠️ The Nex team explicitly recommends against modifying the chat template. The model was trained strictly on the original template — deviating from the training-time format may degrade output quality. See discussion #3 for details.

The recommended approach is to use Nex's patched llama.cpp with the unmodified GGUF. Once Nex's upstream patch is merged into stock llama.cpp, these template-fixed GGUFs will be superseded.

Use this only if you cannot use Nex's patched llama.cpp and need thinking mode to work on stock builds. Be aware that output quality may differ from the original model.

💡 What is APEX?

These GGUF files are quantized using APEX, a MoE-aware mixed-precision quantization technique that outperforms standard quantization methods while being significantly smaller.

APEX beats Q8_0 perplexity at half the size — and even beats F16.

APEX classifies every tensor by its role — routed expert, shared expert, or attention — and applies a layer-wise precision gradient, giving the most sensitive edge layers higher precision and compressing the redundant middle layers more aggressively.

📦 Available Files

File	Size	BPW	Note
`Nex-N2-mini.BF16.gguf`	64.6 GB	16.0	Full precision reference
`Nex-N2-mini-APEX-Quality.gguf`	21.3 GB	5.23	Highest quality, best accuracy
`Nex-N2-mini-APEX-Balanced.gguf`	23.6 GB	5.85	Best all-rounder, recommended
`Nex-N2-mini-APEX-Compact.gguf`	15.4 GB	3.81	Best quality/size ratio, 16GB VRAM
`mmproj-Nex-N2-mini.F16.gguf`	858 MB	-	Vision projector (required for image/video)
`original-chat-template.jinja`	7.9 KB	-	Original unmodified template — for reference / use with Nex's patched llama.cpp

⚠ All GGUF files above (except mmproj) contain a modified chat_template.jinja. See warning above.

🧠 Model Details

Architecture	Qwen3.5 MoE (GatedDeltaNet + Full Attention) + Vision Encoder
Parameters	35B total, 3B active per token
Experts	256 routed experts, 8 active per token
Layers	40 layers (30 linear_attn + 10 full_attn)
Context	262,144 tokens
Vision	Image support (mmproj 858MB)
Thinking	Qwen3-style think tags — works on stock llama.cpp via modified template

🚀 Usage (Stock llama.cpp)

Text only

./llama-server \ -m Nex-N2-mini-APEX-Quality.gguf \ -ngl 99 -ncmoe 19 -c 32768 \ --host 0.0.0.0 --port 8081

With vision

./llama-server \ -m Nex-N2-mini-APEX-Quality.gguf \ --mmproj mmproj-Nex-N2-mini.F16.gguf \ -ngl 99 -ncmoe 19 -c 32768 \ --host 0.0.0.0 --port 8081

No --chat-template-file needed — the fixed template is embedded in the GGUF. Thinking mode works out of the box. Add --mmproj mmproj-Nex-N2-mini.F16.gguf for vision. Replace Nex-N2-mini-APEX-Quality.gguf with your preferred quantization tier (Quality / Balanced / Compact). Recommended sampling: temperature 0.7, top_p 0.95, top_k 40, min_p 0.

📋 Original Model Benchmarks

Benchmark	Score	Category
BrowseComp	74.1	Agent
SWE-Bench Verified	74.4	Coding
Terminal-Bench 2.1	60.7	Coding
GPQA Diamond	82.6	Reasoning
IFEval	89.1	Instruction

From the original Nex-N2-mini model card (BF16, full precision).

SC117/Nex-N2-mini-template-fix-APEX-GGUF

Nex-N2-mini

Links