RoPE-free long-range attention, trained at 4k with zero-shot length extrapolation. FS scores all completed chunk mean keys and reads four selected chunks per query and head. This 341M research base model was trained on FineWeb-Edu at a 4,096-token context. The short retrieval example asks for the hidden number 5437923. Start with CUDA-enabled PyTorch; Usage gives the tested installation and speed-replication command. FreeSolo learns simple chunk mean keys that support search, norm pruning, and hierarchical summarization. RoPE stays in the local window; long-range attention reads selected original tokens without rotating their queries and keys. 12 tasks at 2k/4k; 13 at 8k/16k, where QA-2 is…
FreeSolo-L-341M is an open-weight model for text generation from Zekun Shi, released under MIT License. It has 341M parameters. At 16-bit it needs about 0.8 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index.
RoPE-free long-range attention, trained at 4k with zero-shot length extrapolation. FS-L retains a bounded pool of chunks by key norm and recency, then reads four selected chunks. Its attention scales linearly with sequence length.
Parameters341M
Context—
Weights1.4 GB
Licensemit
AccessOpen weights
Monthly Downloads—
Runs On
What it takes to serve FreeSolo-L-341M (341M parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 0.7 GB | 0.8 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 0.3 GB | 0.4 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 0.2 GB | 0.2 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 16, 2026.
FreeSolo-L-341M on every accelerator the SAVRN Index prices, at every precision
Model Card
By Zekun Shi, published under mit, revision 2d9af652e0de.
RoPE-free long-range attention, trained at 4k with zero-shot length extrapolation. FS-L retains a bounded pool of chunks by key norm and recency, then reads four selected chunks. Its attention scales linearly with sequence length. This 341M research base model was trained on FineWeb-Edu at a 4,096-token context. The short retrieval example asks for the hidden number 5437923. Start with CUDA-enabled PyTorch; Usage gives the tested installation and speed-replication command. FreeSolo learns simple chunk mean keys that support search, norm pruning, and hierarchical summarization. RoPE stays in the local window; long-range attention reads selected original tokens without rotating their queries…
Read Zekun Shi's full model card
Paper · GitHub · Models · Usage
RoPE-free long-range attention, trained at 4k with zero-shot length extrapolation.
FS-L retains a bounded pool of chunks by key norm and recency, then reads four selected chunks. Its attention scales linearly with sequence length. This 341M research base model was trained on FineWeb-Edu at a 4,096-token context.
Try it
hf download zekun-shi/FreeSolo-L-341M freesolo-source.zip --local-dir ./FreeSolo-L-341M
python -m pip install ./FreeSolo-L-341M/freesolo-source.zip
freesolo-generate --model zekun-shi/FreeSolo-L-341M --output generation.json
The short retrieval example asks for the hidden number 5437923. Start with CUDA-enabled PyTorch; Usage gives the tested installation and speed-replication command.
The idea
FreeSolo learns simple chunk mean keys that support search, norm pruning, and hierarchical summarization. RoPE stays in the local window; long-range attention reads selected original tokens without rotating their queries and keys.
Results
| Context | 2k | 4k | 8k | 16k |
|---|---|---|---|---|
| RULER mean | 32.32 | 21.72 | 15.27 | 10.36 |
12 tasks at 2k/4k; 13 at 8k/16k, where QA-2 is available. 100 examples per task, seed 42. These are checkpoint evaluation results; the figures show the paper's model-family comparison.
This is a base model without instruction tuning. Retrieval is task dependent, especially when several facts or values must be combined. Training details, artifact checksums, and citation are in Usage.
Configuration
- Vocabulary size
- 32,000
- Model type
- freesolo
Identity and Version
- Repository
- zekun-shi/FreeSolo-L-341M
- Publisher
- Zekun Shi
- Task
- Text generation
- Modality
- Text
- Library
- pytorch
- Parameters
- 341M parameters
- Languages
- en
- Revision
- 2d9af652e0de6ff81a2c3e9f373e1a7deb0a4543
- First published
- 2026-10-15
- Last updated
- 2026-10-16
Files and Weights
10 files, 1.4 GB in total. The weights are 1 file totalling 1.4 GB in safetensors.
Weights1 file · 1.4 GB
Configuration1 file · 1.5 KB
Documentation3 files · 8.2 KB
Other4 files · 1.7 MB
Repository1 file · 1.7 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.safetensors | Weights | 1.4 GB | a14b3cbfedd1 |
| config.json | Configuration | 1.5 KB | — |
| LICENSE | Documentation | 1.1 KB | — |
| README.md | Documentation | 2.3 KB | — |
| USAGE.md | Documentation | 4.8 KB | — |
| assets/figure-1.png | Other | 122.7 KB | f1937f7b6ee3 |
| assets/figure-2.png | Other | 296.9 KB | 192b3bd396e9 |
| assets/figure-3.png | Other | 254.8 KB | 055de9b7fd7f |
| freesolo-source.zip | Other | 987.5 KB | 7d3c4c6a9174 |
| .gitattributes | Repository | 1.7 KB | — |
License and Download
- License
- mit
- Access
- Open weights, no gate
- Download size
- 1.4 GB
Download from Zekun Shi
Released by Zekun Shi through its official repository on Hugging Face. Read the license.
Built From
- Derived from zekun-shi/FreeSolo-341M
- Described by arXiv:2610.17191
- Trained on (disclosed) HuggingFaceFW/fineweb-edu
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 1.4 GB |
| 16-bit | 0.7 GB |
| 8-bit | 0.3 GB |
| 4-bit | 0.2 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Built on This Model
- Derived fromFreeSolo-B-341M
Questions About FreeSolo-L-341M
How much GPU memory does FreeSolo-L-341M need?
About 0.8 GB at 16-bit and 0.2 GB at 4-bit: the weights (341M parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run FreeSolo-L-341M on?
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use FreeSolo-L-341M commercially?
Yes. FreeSolo-L-341M is released under MIT License. The MIT License is a short permissive license. It permits commercial use, modification and redistribution, provided the copyright notice and permission notice are included.
Similar Models
This is the model card of a transformers model that has been pushed on the Hub. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).
Suggest a title and description for any text. On-device titles and descriptions: a short factual title and a one- to two-sentence description for any passage of text. Swift (requirements) Then add the Title product to your target. The MLX trait is required: without it the module compiles as a stub. Get a title and a one or two sentence description for any passage of text, on device. Fine-tuned on transcript clips, but it works on any prose. The register is deliberately plain, with no emoji, no hashtags and no clickbait, and a description is meant to identify this passage rather than its topic. An MLX model directory. Load the folder, not a single file. The chat template is not incidental. A…
This is the model card of a transformers model that has been pushed on the Hub. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).
English | MiniCPM5-2B-DSpark is a DSpark draft checkpoint trained for exact pairing with MiniCPM5-2B and its tokenizer. Training responses were generated by MiniCPM5-2B from a mixture of general-domain, mathematics, and code prompts. Acceptance length is defined as the total number of completion tokens divided by the total number of speculative verification steps. Evaluation uses natural EOS termination and maxnewtokens=4096. This model has no autonomous intent or legal personhood; its outputs are text generated from statistical patterns and may be inaccurate, biased, or offensive, and may be manipulated by carefully crafted prompts ("jailbreaks") into producing unintended content. Its…
Sanskrit-322M-Base is a 322-million-parameter Sanskrit language model trained from scratch with its own 32,000-token SentencePiece Unigram tokenizer. It is intended for Sanskrit text completion, generation experiments, and further task-specific fine-tuning. This is a base pretrained model: it predicts the continuation of text. It has not been instruction-tuned or trained as a conversational assistant. It uses a Llama-compatible decoder architecture; its weights were initialized and trained independently, rather than adapted from a pretrained Llama checkpoint. Published by sraivante under the Apache License 2.0. The tokenizer uses identity normalization, byte fallback, and SentencePiece…