Llama 3.2 3B Instruct (Q4_K_M)¶

Model card

As an immersion guard, blocks 32 in 100 general attacks and 64 in 100 world attacks at a 5% false-block budget. Writes 256 tokens per second and needs 2.6 GB of VRAM at 8K context.

Llama · 3B · Q4_K_M · 2.0 GB file · bartowski/Llama-3.2-3B-Instruct-GGUF

_{Measured 2026-10-07 · Tryll 0.8.0-DEV · llama.cpp b10839 · NVIDIA GeForce RTX 5070 Ti · vulkan · one deterministic pass (classifiers do not sample) · speed: median of 10 runs · ranked against 13 models}

Plain words¶

What the measured numbers mean for a game. Intervals and ranks are under In the field.

Classification · General guard

32 in 100 · Classification · General guard. Blocks 32 in 100 attempts to pull the NPC out of the game: asking whether it is an AI, jailbreaks, prompt leaks, coding requests and format tricks. Normal player lines wrongly blocked: 1 in 100 on held-out lines (the budget is 5 in 100, set on a separate tuning set).

Classification · World guard

64 in 100 · Classification · World guard. Blocks 64 in 100 lines that do not belong in the setting, such as paying by phone in a medieval town or asking how long chicken keeps in the fridge. Normal player lines wrongly blocked: 3 in 100 on held-out lines (the budget is 5 in 100, set on a separate tuning set).

World guard is measured on one setting, a medieval merchant. A different world needs its own check.

9,133 tokens/s · Performance · Prefill. Reads a 3,974-token prompt in 0.44 s. The first token reaches the player after 0.45 s. A 1,163-token prompt: first token after 0.14 s.

256 tokens/s · Performance · Generation. Writes 256 tokens per second after a 1K-token prompt, and 236 after a 4K-token prompt. 200 tokens take 0.8 s.

2.6 GB VRAM · Performance · Memory. Weights 2.01 GB, KV cache 0.50 GB for 8,192 tokens (q8_0), compute buffers 0.07 GB. Each extra agent on this model adds its own KV cache and compute buffers; the weights are shared.

Speed and memory were measured on one GPU, named in the header. Other hardware gives other numbers.

Not measured yet¶

NPC roleplay: clean speech, staying in lore and role, writing quality
Tool calling: call accuracy and restraint
Summarization: how long the summary keeps what the player said
Intent classification: the butler set is too small for a card number

In the field¶

Where this model sits among the measured models. Right is better. The blue bar is the 95% interval. Guard ranks come from paired comparisons; a speed rank is above another only when the two intervals do not overlap. Close models share a range.

General guard · #7–12 of 13 · setting-invariant classifier

General guard