Skip to main content
Nebius Token Factory supports fine-tuning on multiple open-weight model families.
This page lists:
  • Which base models you can fine-tune
  • Which context lengths they support
  • Which fine-tuning types are available (LoRA vs full fine-tuning)
Deployment note
Not all models that can be fine-tuned can be deployed as serverless endpoints in Nebius Token Factory.

For serving options, see Deploy custom model and the list of available deployment models.

Model List

For each models listed below, Nebius Token Factory supports the following

context_length: 8192, 16384, 32768, 65536, 131072
Unless you override it via the context_length hyperparameter, the default context length for fine-tuning is 8192 tokens. Check hyperparameter section for model details regarding context_length

Meta (Llama 3.1 / 3.2 / 3.3)

Nebius Token Factory and the Meta models hosted in the service are built on the Llama 3.1, Llama 3.2, and Llama 3.3 families. For acceptable use, see Meta’s policies:
NameSupported fine-tuning typeModel card / license
meta-llama/Meta-Llama-3.1-8B-Instruct
(Model card)
LoRA and full fine-tuningLlama 3.1 Community License Agreement
meta-llama/Meta-Llama-3.1-8B
(Model card)
LoRA and full fine-tuningLlama 3.1 Community License Agreement
meta-llama/Llama-3.1-70B-Instruct
(Model card)
LoRA and full fine-tuningLlama 3.1 Community License Agreement
meta-llama/Llama-3.1-70B
(Model card)
LoRA and full fine-tuningLlama 3.1 Community License Agreement
meta-llama/Llama-3.2-1B-Instruct
(Model card)
LoRA and full fine-tuningLlama 3.2 Community License Agreement
meta-llama/Llama-3.2-1B
(Model card)
LoRA and full fine-tuningLlama 3.2 Community License Agreement
meta-llama/Llama-3.2-3B-Instruct
(Model card)
LoRA and full fine-tuningLlama 3.2 Community License Agreement
meta-llama/Llama-3.2-3B
(Model card)
LoRA and full fine-tuningLlama 3.2 Community License Agreement
meta-llama/Llama-3.3-70B-Instruct
(Model card)
LoRA and full fine-tuningLlama 3.3 Community License Agreement

Qwen

Nebius Token Factory supports both dense and coder variants across Qwen3 and Qwen2.5 families.
All Qwen models below use the Apache 2.0 license (see each model card for details).

Qwen3 dense + base models

NameSupported fine-tuning typeModel card / license
Qwen/Qwen3-32B
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen3-14B
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen3-14B-Base
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen3-8B
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen3-8B-Base
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen3-4B
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen3-4B-Base
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen3-1.7B
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen3-1.7B-Base
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen3-0.6B
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen3-0.6B-Base
(Model card)
LoRA and full fine-tuningApache 2.0

Qwen3 coder models

NameSupported fine-tuning typeModel card / license
Qwen/Qwen3-Coder-30B-A3B-Instruct
(Model card)
Full fine-tuningApache 2.0
Qwen/Qwen3-Coder-480B-A35B-Instruct
(Model card)
Full fine-tuningApache 2.0

Qwen2.5 dense + coder models

NameSupported fine-tuning typeModel card / license
Qwen/Qwen2.5-0.5B
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen2.5-0.5B-Instruct
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen2.5-7B
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen2.5-7B-Instruct
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen2.5-14B
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen2.5-14B-Instruct
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen2.5-32B
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen2.5-32B-Instruct
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen2.5-72B
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen2.5-72B-Instruct
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen2.5-Coder-32B
(Model card)
LoRA and full fine-tuningApache 2.0
Qwen/Qwen2.5-Coder-32B-Instruct
(Model card)
LoRA and full fine-tuningApache 2.0

OpenAI / Unsloth GPT-OSS

These models are OpenAI GPT-OSS weights (bf16) packaged by Unsloth.
They are Apache 2.0–licensed and suitable for both research and commercial use (subject to the license).
To convert the weights into MXFP4 please follow instructions here.
NameSupported fine-tuning typeModel card / license
unsloth/gpt-oss-20b-BF16
(Model card)
LoRA and full fine-tuningApache 2.0
unsloth/gpt-oss-120b-BF16
(Model card)
LoRA and full fine-tuningApache 2.0
For merging MoE LoRA adapter weights please follow the guide here.

DeepSeek

Nebius Token Factory integrates DeepSeek V3 models for high-capacity reasoning workloads.
DeepSeek V3 and its variants are released under the MIT License (see model cards for details).
NameSupported fine-tuning typeModel card / license
deepseek-ai/DeepSeek-V3-0324
(Model card)
Full fine-tuningMIT License
deepseek-ai/DeepSeek-V3.1
(Model card)
Full fine-tuningMIT License
DeepSeek V3 and DeepSeek V3.1 are currently only available in Nebius US data centers.

Base LoRA adapter models available for deployment

You can deploy serverless LoRA adapter models in Nebius Token Factory with per-token billing.
To deploy a LoRA-adapted model, first fine-tune an adapter on one of the base models below:
NameSupported fine-tuning type for adaptersLicense
meta-llama/Meta-Llama-3.1-8B-Instruct
(Model card)
LoRA and full fine-tuning (LoRA deployable as serverless)Llama 3.1 Community License Agreement
meta-llama/Llama-3.3-70B-Instruct
(Model card)
LoRA fine-tuning (LoRA deployable as serverless)Llama 3.3 Community License Agreement
For other models listed on this page, fine-tuning is supported, but deployment options may differ (for example, only via custom hosting or future releases of serverless runtimes).