Qwen2.5-7B-Instruct-Helix

2.2x smaller from BF16. Beats GPTQ. Zero calibration data.

Qwen2.5-7B-Instruct compressed from 14.2 GB (BF16) to 6.5 GB. Beats GPTQ quality (+6.34% vs +8.2% PPL) and AWQ (+11.1%) with zero calibration data. No fine-tuning. Just pip install and from_pretrained().

Install and Run

pip install "helix-substrate[hf]"

import helix_substrate  # registers the helix quantizer with HuggingFace
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("EchoLabs33/qwen2.5-7b-instruct-helix")
tokenizer = AutoTokenizer.from_pretrained("EchoLabs33/qwen2.5-7b-instruct-helix")

inputs = tokenizer("Explain the theory of relativity in simple terms:", return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

That's it. import helix_substrate registers the quantizer. from_pretrained() handles the rest automatically.

Compression Benchmark

	Dense (BF16)	Helix (HXQ)
Size	14.2 GB	6.5 GB
Perplexity (WikiText-2)	6.949	7.388 (+6.34%)
Compression ratio	—	2.2x
Compressed modules	—	196 HelixLinear layers
Architecture	Qwen2 (28 layers, GQA)	unchanged

Eval: WikiText-2 test split, 2048 tokens, stride 512.

Quality vs GPTQ and AWQ

Method	PPL	PPL Delta	Calibration Data
FP16 Dense	6.949	baseline	—
HXQ (HelixLinear k=256)	7.388	+6.34%	None
GPTQ Int4	7.518	+8.2%	128 sequences
AWQ Int4	7.719	+11.1%	Activation stats

HXQ beats GPTQ by 23% less degradation, and beats AWQ by 43% — with zero calibration data.

Good to Know

GPU and CPU supported — runs on any CUDA GPU or CPU via standard PyTorch. Fused kernels for additional speedup are in progress.
Fine-tunable via LoRA — compressed weights remain frozen, but LoRA adapters attach to each HelixLinear layer via HelixLinearSTE. See helix-substrate for training infrastructure.
Requires helix-substrate — the quantizer is not built into transformers. You need pip install "helix-substrate[hf]".
Tied embeddings — lm_head shares embed_tokens, stored at full precision.
Requires 8+ GB VRAM — fits on RTX 3060/4060 or higher.

What is HelixCode?

HelixCode is a universal weight compression codec based on vector quantization:

Each weight matrix is replaced by a 256-entry codebook (float32) + uint8 index matrix + optional sidecar corrections for outlier values
The compressed form is the executable — HelixLinear performs codebook[indices] @ x directly, no decompression step
Works on any nn.Linear regardless of architecture (Transformer, Mamba, MLP, CNN)
No calibration data required — unlike GPTQ/AWQ, codebooks are fit from the weights alone

How It Works

import helix_substrate registers the hxq quantizer with HuggingFace
from_pretrained() reads quantization_config.quant_method = "hxq" from config.json
The quantizer replaces 196 nn.Linear modules with HelixLinear shells before weight loading
Safetensors populates the codebook, indices, and sidecar buffers directly
The model runs in compressed form — no decompression needed

Compression Receipt

Compressed tensors:  196
Exact tensors:       143  (norms, embeddings)
Dense size:          14.2 GB (BF16)
Compressed size:     6.5 GB
Compression ratio:   2.2x
PPL delta:           +6.34% (7.388 vs 6.949 dense)
Eval: WikiText-2 test, 2048 tokens, stride=512

Companion Models

Same codec, same pip install, multiple architectures:

Model	Architecture	Ratio	PPL Delta
qwen2.5-14b-instruct-helix	Transformer	3.4x	pending
qwen2.5-3b-instruct-helix	Transformer	1.6x	+0.69%
qwen2.5-coder-3b-helix	Transformer (code)	1.6x	+1.92%
qwen2.5-coder-1.5b-instruct-helix	Transformer (code)	2.4x	+1.63%
tinyllama-1.1b-helix	Transformer	4.0x	+0.78%
zamba2-2.7b-instruct-helix	Hybrid (Mamba2+Transformer)	1.8x	+6.59%
zamba2-1.2b-helix	Hybrid (Mamba2+Transformer)	1.7x	+2.90%
mamba2-1.3b-helix	Pure SSM (Mamba2)	2.1x	+8.0%
mamba-130m-helix	Pure SSM	3.8x	+18.4%

Citation

@software{helix_substrate_2026,
  title={Helix Substrate: Universal Weight Compression via HelixCode},
  author={EchoLabs},
  year={2026},
  url={https://github.com/echo313unfolding/helix-substrate}
}