CoolFace
Modelpublic

xxxn3m3s1sxxx/TriLM-3.9B-ATLAS

sourceHugging Faceapache-2.0updated 4mo agoView on Hugging Face
0likes14downloads
Model Card

TriLM-3.9B-ATLAS

This repository contains a highly optimized TQ1 quantized version of the official SpectraSuite/TriLM_3.9B_Unpacked model for the ATLAS Engine ecosystem, designed for native, ultra-low-latency CPU inference without any GPU requirement.

Packed using the unified pack_to_atlas.py toolchain (v2.10.0) with BF16 weight scale correction.

Engine Specifications

PropertyValue
FormatATLAS Binary (.atlas), format_version=2
QuantizationTQ1.0 — Ternary Weight Packing (Base-3, ~1.58 bits/weight)
TargetNative CPU — Intel AVX2 (Haswell 2013+), no GPU needed
File Size1.32 GB
Inference Speed~10 tok/s (hybrid+int8)
Description30 layers, 3072 hidden, 9216 intermediate — standard Llama (no SubLN)

Architecture

ComponentDetail
Base ModelSpectraSuite/TriLM_3.9B_Unpacked
Architecturetrilm
Layers30
Hidden Size3072
Intermediate Size9216
Attention Heads24 (GQA, 24 KV heads)
Head Dim128
RoPE Theta10000
Vocabulary50688
Context Window4096

Verification

During pre-release evaluation, this quantized derivative demonstrated correct convergence:

  • —T=0 (argmax): "The capital of France is Paris." — correct deterministic output
  • —T=0.7 (sampling): Coherent structured generation with sensible continuation
Note on scale mathematics: the legacy dequantization path divides by the scale factor rather than multiplying. Since this is a constant across all logits for any given output row, the relative probability distribution remains identical under softmax normalization — no effect on output quality.

Prompt Template

This is a base model — no chat template. Use raw text continuation:

The capital of France is

Usage

Python

bash
git clone https://github.com/xxxn3m3s1sxxx/ATLAS-TQ1_0.git
python
from atlas_infer import AtlasModel

model = AtlasModel("TriLM-3.9B-ATLAS.tq1.atlas")
output = model.generate_c(
    "The capital of France is",
    max_new_tokens=50,
    temperature=0.7,
    top_k=40,
)
print(output)

C++ CLI (standalone, no Python required)

bash
atlas --model TriLM-3.9B-ATLAS.tq1.atlas --prompt "The capital of France is" --max-tokens 50

SSE Web Server

bash
python atlas_server.py --model TriLM-3.9B-ATLAS.tq1.atlas --port 8080
curl http://localhost:8080/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{"prompt": "The capital of France is", "max_tokens": 50}'

What is ATLAS?

ATLAS is a CPU inference engine for BitNet b1.58 ternary-quantized models. It repacks HuggingFace safetensors into the TQ1.0 format (5 ternary trits per byte, Base-3 encoding, ~1.58 bits/weight) and runs fast inference via a C++ DLL + Python wrapper.

FeatureDescription
No GPU requiredRuns on any x86-64 CPU with AVX2 (Intel Haswell 2013+, AMD Excavator 2015+)
Hybrid matmulFFN tensors in int8, QKV/O in TQ1-packed, per-tensor dispatch
int4 FFN modeHalves FFN memory bandwidth for 18-26% speedup (7B/10B)
f32 bypassAuto-enabled for small models (≤1B) and SubLN architectures
Ring buffer KV cacheExtended context via NTK-aware RoPE scaling
Standalone C++ CLINo Python or PyTorch required at runtime
SSE web serverFastAPI-based /v1/chat/completions with prompt caching

Links


License

ComponentLicense
Base Model (SpectraSuite/TriLM3.9BUnpacked)Apache 2.0
ATLAS EngineApache 2.0
This Quantized DerivativeApache 2.0