microsoft/bitnet-b1.58-2B-4T-bf16

BitNet b1.58 2B4T - Scaling Native 1-bit LLM

This repository contains the weights for BitNet b1.58 2B4T, the first open-source, native 1-bit Large Language Model (LLM) at the 2-billion parameter scale, developed by Microsoft Research.

Trained on a corpus of 4 trillion tokens, this model demonstrates that native 1-bit LLMs can achieve performance comparable to leading open-weight, full-precision models of similar size, while offering substantial advantages in computational efficiency (memory, energy, latency).

➡️ Technical Report: BitNet b1.58 2B4T Technical Report

➡️ Official Inference Code: microsoft/BitNet (bitnet.cpp)

Model Variants

Several versions of the model weights are available on Hugging Face:

• microsoft/bitnet-b1.58-2B-4T: Contains the packed 1.58-bit weights optimized for efficient inference. Use this for deployment.

• microsoft/bitnet-b1.58-2B-4T-bf16 (This repository): Contains the master weights in BF16 format. Use this only for training or fine-tuning purposes.

• microsoft/bitnet-b1.58-2B-4T-gguf: Contains the model weights in GGUF format, compatible with the bitnet.cpp library for CPU inference.

Model Details

• Architecture: Transformer-based, modified with BitLinear layers (BitNet framework).
  • Uses Rotary Position Embeddings (RoPE).
  • Uses squared ReLU (ReLU²) activation in FFN layers.
  • Employs subln normalization.
  • No bias terms in linear or normalization layers.
• Quantization: Native 1.58-bit weights and 8-bit activations (W1.58A8).
  • Weights are quantized to ternary values {-1, 0, +1} using absmean quantization during the forward pass.
  • Activations are quantized to 8-bit integers using absmax quantization (per-token).
  • Crucially, the model was trained from scratch with this quantization scheme, not post-training quantized.
• Parameters: ~2 Billion
• Training Tokens: 4 Trillion
• Context Length: Maximum sequence length of 4096 tokens.
  • Recommendation: For optimal performance on tasks requiring very long contexts (beyond the pre-training length or for specialized long-reasoning tasks), we recommend performing intermediate long-sequence adaptation/training before the final fine-tuning stage.
• Training Stages:
  1. Pre-training: Large-scale training on public text/code and synthetic math data using a two-stage learning rate and weight decay schedule.
  2. Supervised Fine-tuning (SFT): Fine-tuned on instruction-following and conversational datasets using sum loss aggregation and specific hyperparameter tuning.
  3. Direct Preference Optimization (DPO): Aligned with human preferences using preference pairs.
• Tokenizer: LLaMA 3 Tokenizer (vocab size: 128,256).

How to Use (with transformers)

VERY IMPORTANT NOTE ON EFFICIENCY

Please do NOT expect performance efficiency gains (in terms of speed, latency, or energy consumption) when using this model with the standard transformers library, even with the required fork.

The current execution paths within transformers do not contain the specialized, highly optimized computational kernels required to leverage the advantages of the BitNet architecture. Running the model via transformers will likely result in inference speeds and energy usage comparable to, or potentially worse than, standard full-precision models within this framework on both CPU and GPU.

While you might observe reduced memory usage due to the quantized weights, the primary computational efficiency benefits are not accessible through this standard transformers usage path.

For achieving the efficiency benefits demonstrated in the technical paper, you MUST use the dedicated C++ implementation: bitnet.cpp.

Requirements

pip install git+https://github.com/shumingma/transformers.git

We are actively working with the Hugging Face team to integrate the necessary code into the main transformers library. This installation method may change in the future.

Example

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "microsoft/bitnet-b1.58-2B-4T"

# Load tokenizer and model
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16
)

# Apply the chat template
messages = [
    {"role": "system", "content": "You are a helpful AI assistant."},
    {"role": "user", "content": "How are you?"},
]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
chat_input = tokenizer(prompt, return_tensors="pt").to(model.device)

# Generate response
chat_outputs = model.generate(**chat_input, max_new_tokens=50)
response = tokenizer.decode(chat_outputs[0][chat_input['input_ids'].shape[-1]:], skip_special_tokens=True) # Decode only the response part
print("\nAssistant Response:", response)

How to Use (with bitnet.cpp)

Please refer to the bitnet.cpp GitHub repository for detailed compilation steps, usage examples, and command-line options.

Evaluation

BitNet b1.58 2B4T was evaluated against leading open-weight full-precision LLMs of similar size. Below are the key results (all models are instruction-tuned versions):

Benchmark	LLaMA 3.2 1B	Gemma-3 1B	Qwen2.5 1.5B	SmolLM2 1.7B	MiniCPM 2B	BitNet b1.58 2B
Memory (Non-emb)	2GB	1.4GB	2.6GB	3.2GB	4.8GB	0.4GB
Latency (CPU Decoding)	48ms	41ms	65ms	67ms	124ms	29ms
Energy (Estimated)	0.258J	0.186J	0.347J	0.425J	0.649J	0.028J
Training Tokens (Pre-train)	9T*	2T**	18T	11T	1.1T	4T
ARC-Challenge	37.80	38.40	46.67	43.52	44.80	49.91
ARC-Easy	63.17	63.13	76.01	62.92	72.14	74.79
OpenbookQA	34.80	38.80	40.80	46.00	40.20	41.60
BoolQ	64.65	74.22	78.04	75.78	80.67	80.18
HellaSwag	60.80	57.69	68.28	71.71	70.81	68.44
PIQA	74.21	71.93	76.12	76.12	76.66	77.09
WinoGrande	59.51	58.48	62.83	68.98	61.80	71.90
CommonsenseQA	58.48	42.10	76.41	63.55	71.74	71.58
TruthfulQA	43.80	38.66	46.67	39.90	41.41	45.31
TriviaQA	37.60	23.49	38.37	45.97	34.13	33.57
MMLU	45.58	39.91	60.25	49.24	51.82	53.17
HumanEval+	31.10	37.20	50.60	28.00	43.90	38.40
GSM8K	38.21	31.16	56.79	45.11	4.40	58.38
MATH-500	23.00	42.00	53.00	17.60	14.80	43.40
IFEval	62.71	66.67	50.12	57.91	36.81	53.48
MT-bench	5.43	6.40	6.12	5.50	6.57	5.85
Average	44.90	43.74	55.23	48.70	42.05	54.19

*LLaMA 3.2 1B uses pruning & distillation.

**Gemma-3 1B uses distillation.

License

The model weights and code are released under the MIT License.

Disclaimer

This model is intended for research and development purposes. While efforts have been made to align it using SFT and DPO, it may still produce outputs that are unexpected, biased, or inaccurate. Please use responsibly.