File size: 14,352 Bytes
b5f4a54 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 | ---
license: mit
language:
- ru
- en
pipeline_tag: text-generation
tags:
- dllm
- diffusion
- diffusion-language-modeling
- instruct
library_name: transformers
---
# Zarya-1.7B
Zarya is a family of hybrid language models that combine a classic auto-regressive (AR) objective with a masked-diffusion (MDM) objective in one model.
The architecture can be built on top of any autoregressive model but in this repository it uses the `Qwen3` backbone.
Naming explanation: Zarya (pronounced as [zɐˈrʲa] ([IPA notation](https://en.wiktionary.org/wiki/Appendix:Russian_pronunciation)), literally "Dawn" in English) is a figure from Slavic folklore — a female personification of dawn who may be considered a goddess.
In various traditions, she can manifest as a single being or as two or three sisters simultaneously.
This is a research prototype.
## Model Details
### Model Description
Zarya is a research prototype of a family of hybrid language models that jointly learn a classic auto-regressive (AR) objective and a masked-diffusion (MDM) objective within a single model.
Two generation modes are supported, both reachable through a single `model.generate(...)` call: masked-diffusion (MDM) sampling and slotted-level speculative parallel decoding.
- **Model type:** Hybrid auto-regressive (AR) + masked-diffusion language model (DLLM); backbone `Qwen3`, wrapper `Zarya`
- **Language(s) (NLP):** Russian and English
- **License:** MIT
- **Preprint:** https://arxiv.org/abs/2609.19868
- **Repository with training code:** https://github.com/ai-forever/zarya
### Zarya-1.7B details
Zarya-1.7B has the following features:
| Variant | hidden_size | num_hidden_layers | num_attention_heads | intermediate_size |
|------------|-------------|-------------------|---------------------|-------------------|
| Zarya-1.7B | 2048 | 28 | 16 | 6144 |
Context Length: 2048
## Uses
Zarya is intended for text generation.
It supports conversational fine-tuning (SFT) and classic auto-regressive pretraining.
### Direct Use
Direct use is text generation (continuation of a prompt) through the `model.generate(...)` interface, including chat-style prompts formatted with the provided chat template.
Two inference modes are available through the same `generate()` call.
Both modes fully use the KV cache with causal attention masks.
- **MDM sampling** (`slotted_generation=false`): iterative masked-diffusion denoising with the first-hitting sampler.
- **Slotted speculative decoding** (`slotted_generation=true`): parallel slot generation with inter-slot diffusion-based selection and intra-slot autoregressive generation for a decoding speedup.
### Out-of-Scope Use
The model is a research prototype.
It should not be used for production decisions, safety-critical applications, or any use case where accuracy and reliability are essential without additional evaluation and safeguards.
Inference performance and stability also depend on the chosen decoding hyperparameters (like `slotted_generation`, `slot_size`, `serial_num_blocks`, `slot_threshold`, `token_threshold`, and others).
## Bias, Risks, and Limitations
This is a research prototype.
The code relies on Hugging Face Transformers APIs; when upgrading versions, compatibility must be checked (tested on Transformers 5.12.1 and PyTorch 2.9.0).
Inference performance and stability depend on the choice of config parameters.
## How to Get Started with the Model
Use the code below to get started with the model. Loading the model and tokenizer requires `trust_remote_code=True`.
```python
import torch
from transformers import AutoModel, AutoTokenizer
model_name = "ai-forever/Zarya-1.7B"
model = AutoModel.from_pretrained(model_name, trust_remote_code=True, torch_dtype=torch.bfloat16).cuda()
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
prompt = "<|im_start|>user\nHello!<|im_end|>\n<|im_start|>assistant\n"
input_ids = tokenizer(prompt, return_tensors="pt").input_ids.cuda()
# Both modes go through model.generate(...).
# With generation_config.slotted_generation=true -> slotted speculative decoding:
out = model.generate(
input_ids,
max_new_tokens=256,
do_sample=True,
temperature=0.7,
slot_size=16,
serial_num_blocks=4,
slot_threshold=0.9,
token_threshold=0.3,
)
# Setting generation_config.slotted_generation=false -> MDM sampling instead:
# out = model.generate(input_ids, max_new_tokens=256)
print(tokenizer.decode(out[0]))
```
## Evaluation
LM-eval benchmarking with the `lm-eval` package is supported. Example run:
```bash
lm_eval run \
--tasks=gsm8k,ifeval,mbpp,mbpp_instruct,mbpp_plus,mbpp_plus_instruct,hellaswag \
--model=hf --confirm_run_unsafe_code \
--log_samples \
--apply_chat_template \
--output_path=./reports/lm-eval_results \
--model_args=pretrained=ai-forever/Zarya-1.7B,backend=causal,dtype=bfloat16,attn_implementation=sdpa,trust_remote_code=True \
--gen_kwargs slotted_generation=true,slot_size=16,serial_num_blocks=4,slot_threshold=0.9,token_threshold=0.4
```
### Zarya-1.7B results
Measurements below were collected with varying inference parameters and on different GPUs; performance is sensitive to both, so results may differ across configurations and hardware setups.
#### A100, `dtype=bfloat16`, `apply_chat_template`, `slotted_generation=true,slot_size=16,serial_num_blocks=4,slot_threshold=0.9,token_threshold=0.4`
Hardware info:
gpu_driver_cuda_version 13.2;
gpu_driver_version 595.71.05;
Docker info:
Torch: 2.9.0+cu128; Transformers: 5.12.1; CUDNN in torch: 91002;
`lm-eval == 0.4.12`
| Tasks | Version | Filter | n-shot | Metric | | Value | | Stderr |
|-----------|--------:|------------------|-------:|-------------------------|---|-------:|---|--------|
| gsm8k | 3 | flexible-extract | 5 | exact_match | ↑ | 0.4670 | ± | 0.0137 |
| | | strict-match | 5 | exact_match | ↑ | 0.4685 | ± | 0.0137 |
| hellaswag | 1 | none | 0 | acc | ↑ | 0.4224 | ± | 0.0049 |
| | | none | 0 | acc_norm | ↑ | 0.5472 | ± | 0.0050 |
| ifeval | 4 | none | 0 | inst_level_loose_acc | ↑ | 0.5815 | ± | N/A |
| | | none | 0 | inst_level_strict_acc | ↑ | 0.5552 | ± | N/A |
| | | none | 0 | prompt_level_loose_acc | ↑ | 0.4492 | ± | 0.0214 |
| | | none | 0 | prompt_level_strict_acc | ↑ | 0.4288 | ± | 0.0213 |
| mbpp | 1 | none | 3 | pass_at_1 | ↑ | 0.3540 | ± | 0.0214 |
| mbpp_plus | 1 | none | 3 | pass_at_1 | ↑ | 0.4815 | ± | 0.0257 |
#### H100, `dtype=bfloat16`, `apply_chat_template`, `slotted_generation=true,slot_size=16,serial_num_blocks=4,slot_threshold=0.9,token_threshold=0.4`
Hardware info:
gpu_driver_cuda_version 13.0;
gpu_driver_version 580.105.08;
Docker info:
Torch: 2.9.0+cu128; Transformers: 5.12.1; CUDNN in torch: 91002;
`lm-eval == 0.4.12`
| Tasks | Version | Filter | n-shot | Metric | | Value | | Stderr |
|--------------------|--------:|------------------|-------:|-------------------------|---|-------:|---|--------|
| arc_challenge | 1 | none | 0 | acc | ↑ | 0.4437 | ± | 0.0145 |
| | | none | 0 | acc_norm | ↑ | 0.4573 | ± | 0.0146 |
| gsm8k | 3 | flexible-extract | 5 | exact_match | ↑ | 0.0553 | ± | 0.0063 |
| | | strict-match | 5 | exact_match | ↑ | 0.0455 | ± | 0.0057 |
| hellaswag | 1 | none | 0 | acc | ↑ | 0.4227 | ± | 0.0049 |
| | | none | 0 | acc_norm | ↑ | 0.5474 | ± | 0.0050 |
| hendrycks_math500 | 1 | none | 0 | exact_match | ↑ | 0.0040 | ± | 0.0028 |
| humaneval | 1 | create_test | 0 | pass@1 | ↑ | 0.0000 | ± | 0 |
| humaneval_instruct | 4 | create_test | 0 | pass@1 | ↑ | 0.0549 | ± | 0.0178 |
| ifeval | 4 | none | 0 | inst_level_loose_acc | ↑ | 0.4317 | ± | N/A |
| | | none | 0 | inst_level_strict_acc | ↑ | 0.4053 | ± | N/A |
| | | none | 0 | prompt_level_loose_acc | ↑ | 0.3013 | ± | 0.0197 |
| | | none | 0 | prompt_level_strict_acc | ↑ | 0.2791 | ± | 0.0193 |
| mbpp | 1 | none | 3 | pass_at_1 | ↑ | 0.0000 | ± | 0 |
| mbpp_plus | 1 | none | 3 | pass_at_1 | ↑ | 0.0000 | ± | 0 |
#### H100, `dtype=bfloat16`, `apply_chat_template`, `slotted_generation=false,T=0,temperature=0.5,top_p=0.8,do_sample=true,noise_schedule=linear`
Hardware info:
gpu_driver_cuda_version 13.0;
gpu_driver_version 580.105.08;
Docker info:
Torch: 2.9.0+cu128; Transformers: 5.12.1; CUDNN in torch: 91002;
`lm-eval == 0.4.12`
| Tasks | Version | Filter | n-shot | Metric | | Value | | Stderr |
|--------------------|--------:|------------------|-------:|-------------------------|---|-------:|---|--------|
| arc_challenge | 1 | none | 0 | acc | ↑ | 0.4437 | ± | 0.0145 |
| | | none | 0 | acc_norm | ↑ | 0.4573 | ± | 0.0146 |
| gsm8k | 3 | flexible-extract | 5 | exact_match | ↑ | 0.0091 | ± | 0.0026 |
| | | strict-match | 5 | exact_match | ↑ | 0.0000 | ± | 0 |
| hellaswag | 1 | none | 0 | acc | ↑ | 0.4227 | ± | 0.0049 |
| | | none | 0 | acc_norm | ↑ | 0.5474 | ± | 0.0050 |
| hendrycks_math500 | 1 | none | 0 | exact_match | ↑ | 0.0000 | ± | 0 |
| humaneval | 1 | create_test | 0 | pass@1 | ↑ | 0.0000 | ± | 0 |
| humaneval_instruct | 4 | create_test | 0 | pass@1 | ↑ | 0.0000 | ± | 0 |
| ifeval | 4 | none | 0 | inst_level_loose_acc | ↑ | 0.2002 | ± | N/A |
| | | none | 0 | inst_level_strict_acc | ↑ | 0.1655 | ± | N/A |
| | | none | 0 | prompt_level_loose_acc | ↑ | 0.1072 | ± | 0.0133 |
| | | none | 0 | prompt_level_strict_acc | ↑ | 0.0776 | ± | 0.0115 |
| mbpp | 1 | none | 3 | pass_at_1 | ↑ | 0.0000 | ± | 0 |
| mbpp_plus | 1 | none | 3 | pass_at_1 | ↑ | 0.0000 | ± | 0 |
#### H100, `dtype=bfloat16`, `apply_chat_template`, `slotted_generation=true,slot_size=16,serial_num_blocks=4,slot_threshold=0.9,token_threshold=0.4,max_gen_toks=2048`
Hardware info:
gpu_driver_cuda_version 13.0;
gpu_driver_version 580.105.08;
Docker info:
Torch: 2.9.0+cu128; Transformers: 5.12.1; CUDNN in torch: 91002;
`lm-eval == 0.4.12`
| Tasks | Version | Filter | n-shot | Metric | | Value | | Stderr |
|--------------------|--------:|------------------|-------:|-------------------------|---|-------:|---|--------|
| arc_challenge | 1 | none | 0 | acc | ↑ | 0.4437 | ± | 0.0145 |
| | | none | 0 | acc_norm | ↑ | 0.4573 | ± | 0.0146 |
| gsm8k | 3 | flexible-extract | 5 | exact_match | ↑ | 0.0167 | ± | 0.0035 |
| | | strict-match | 5 | exact_match | ↑ | 0.0099 | ± | 0.0027 |
| hellaswag | 1 | none | 0 | acc | ↑ | 0.4227 | ± | 0.0049 |
| | | none | 0 | acc_norm | ↑ | 0.5474 | ± | 0.0050 |
| hendrycks_math500 | 1 | none | 0 | exact_match | ↑ | 0.0000 | ± | 0 |
| humaneval | 1 | create_test | 0 | pass@1 | ↑ | 0.0000 | ± | 0 |
| humaneval_instruct | 4 | create_test | 0 | pass@1 | ↑ | 0.0671 | ± | 0.0196 |
| ifeval | 4 | none | 0 | inst_level_loose_acc | ↑ | 0.4508 | ± | N/A |
| | | none | 0 | inst_level_strict_acc | ↑ | 0.4269 | ± | N/A |
| | | none | 0 | prompt_level_loose_acc | ↑ | 0.3142 | ± | 0.0200 |
| | | none | 0 | prompt_level_strict_acc | ↑ | 0.2865 | ± | 0.0195 |
| mbpp | 1 | none | 3 | pass_at_1 | ↑ | 0.0000 | ± | 0 |
| mbpp_plus | 1 | none | 3 | pass_at_1 | ↑ | 0.0000 | ± | 0 |
---
## Citation
If you find our work helpful, please consider citing (citation will be updated after peer-reviewed publication):
```bibtex
@misc{sinev-etal-2026-Zarya,
author = {Sinev, Leonid and Koziev, Ilya and Leshchuk, Vladislav},
title = {Zarya: A Hybrid Autoregressive--Masked Diffusion Language Model with Flexible Training and Dual-Mode Inference},
year = {2026},
archiveprefix = {arXiv},
eprint = {2609.19868},
primaryclass = {cs.CL},
url = {https://arxiv.org/abs/2609.19868},
}
```
|