Downloads · 30 days
1.2M
73% of all-time downloads
ukisai/Swift-1.5-Qwen3.8-Flash-Next-GSQ-RCO-GGUF
Swift-1.5-Qwen3.8-Flash-Next-GSQ-RCO-GGUF is a image-text-to-text model from ukisai. Use it for the image-text-to-text task on the model card, and read the license before you ship it in a product. It is set up for gguf. The card lists the license as other.
<div align="center" <a href="https://ukisai.com"<img src="ukisai-banner.png" alt="UkisAI" style="width:100%;height:auto;" /</a <p<a href="https://ukisai.com"<bWebsite</b</a • <a href="https://ukisai.com/products/…
Downloads · 30 days
1.2M
73% of all-time downloads
All-time downloads
1.7M
Public
Repo size
380 GB
Likes
273
Trending 20
Click a slice to open those files.
.gguf380 GB · 100%
From the Hugging Face model README
Mixed-precision GGUF quantizations of Swift Flash Next, with Swift-specific GSQ refinement and reused ISTA GSQ-RCO per-tensor allocation profiles.
Swift Flash Next is UkisAI's reasoning-efficient derivative of Qwen3.8-Flash-Next. Its post-training targets shorter reasoning traces and coding, agentic and long-horizon tasks. See the original model card for model-level benchmarks and training details. Those benchmarks are separate from the quantization measurements below.
Swift 1.5 Flash-Next uses 63.4% fewer thinking tokens, with a 1.8x speed up while keeping the accuracy loss <1% vs base on xhigh.
Each tier contains two GGUF shards. Download both files into the same directory and load shard 1; llama.cpp locates shard 2 automatically. Sizes are decimal GB and exclude runtime context/cache memory. Tier names describe mixed-precision allocation profiles.
| Tier | Combined GGUF size | Shards | Development KLD ↓ |
|---|---|---|---|
| IQ3_S_G8 (G8) | 84.42 GB | 1 · 2 | not measured |
| IQ3_S | 83.74 GB | 1 · 2 | 0.144674 |
| IQ3_XXS | 75.97 GB | 1 · 2 | 0.240139 |
| IQ2_XS | 68.15 GB | 1 · 2 | 0.341275 |
| Q2_0 (experimental) | 66.55 GB | 1 · 2 | 0.424350 |
G8 (IQ3_S_G8) is the IQ3_S tier with one change: the 108 weight matrices of the 36 linear-attention (Gated DeltaNet) layers are Q8_0. All other tensors are byte-identical to IQ3_S (+0.67 GB). It is the default file of the Swift engine early access, for 128 GB machines.
A BF16 vision projector is included separately (0.91 GB). The evaluation below covers text inference; it does not measure vision-task accuracy.
Exact model and shard identities are recorded in release-manifest.json and SHA256SUMS.
Original evaluated, unsplit GGUF files can also be reconstructed byte for byte using the exact-source recovery files and script. These small files preserve headers and padding; normal inference needs only the two model shards. All four reconstructed source hashes were verified before release.
KLD measures divergence from the corresponding BF16 model's next-token distribution; lower is better. Swift quants are measured against Swift Flash Next BF16. Development measurements use 100 chunks at a 512-token context, and informed refinement.
Reporting prose, code and math sets use 100 chunks each; German, French, Spanish and Chinese use 25 chunks each, all at context 512. The seven original reporting sets were reused. IQ3_S, IQ2_XS and Q2_0 additionally have results on a preregistered fresh English C4 shard; no matching fresh result is available for IQ3_XXS.
| Reporting text | IQ3_S | IQ3_XXS | IQ2_XS | Q2_0 (experimental) |
|---|---|---|---|---|
| English prose | 0.062556 | 0.116077 | 0.188117 | 0.234242 |
| Fresh English sample | 0.067801 | — | 0.186271 | 0.236228 |
| CodeParrot code | 0.069016 | 0.118707 | 0.174294 | 0.235256 |
| GSM8K math text | 0.054960 | 0.086821 | 0.120991 | 0.149809 |
| German | 0.063634 | 0.109100 | 0.166814 | 0.219444 |
| French | 0.076760 | 0.133624 | 0.213379 | 0.300467 |
| Spanish | 0.042848 | 0.073681 | 0.119114 | 0.148757 |
| Chinese | 0.095915 | 0.174102 | 0.264923 | 0.385924 |
IQ2_XS is the standout. It has lower KLD than ISTA-DASLab's own GSQ-RCO IQ2_XS on seven of eight reporting sets, by 5–11% (math text −8.9%, Chinese −10.5%), and improves on its Swift starting quant in every domain, by 8–17%. IQ3_XXS improves on its Swift starting quant in six of seven domains, by 3–16%, and has lower KLD than ISTA's IQ3_XXS on math text (−7.8%) and Chinese (−5.1%); the other domains are within 1–4%.
[!NOTE] Q2_0 is experimental. It improves on its Swift starting quant in five of seven domains, but has higher KLD than ISTA's Q2_0 on six of eight reporting sets. For a file of similar size, prefer IQ2_XS.
IQ3_S has the lowest KLD of the four tiers. On the development set its KLD is at parity with ISTA-DASLab's GSQ-RCO IQ3_S (0.144674 vs 0.146683, −1.4%). On the eight reporting sets it has lower KLD than ISTA's IQ3_S on 4 (Chinese −7.2%, math text −3.7%, Spanish −0.4%, code −0.1%) and higher KLD on 4 (French +8.8%, German +2.8%, fresh English C4 +0.6%, English prose +0.3%). It improves on its Swift starting quant in 8 of 8 domains, by 6–17%.
These comparisons use each model's own BF16 reference. They are not direct capability rankings, task-accuracy percentages or statistical-equivalence claims.
The full result table includes ISTA and starting-quant comparisons and reported error estimates. Evaluation metadata binds these results to the selected model identities and records limitations. Lexical overlap filtering does not prove semantic deduplication or absence of overfitting. These tests do not establish long-context quality.
Use a llama.cpp build supporting Qwen3.8-Flash-Next.
hf download ukisai/Swift-1.5-Qwen3.8-Flash-Next-GSQ-RCO-GGUF \
--include "Swift-Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-*.gguf" --local-dir .
llama-server \
-m Swift-Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00001-of-00002.gguf \
--jinja -fa on -ngl 99 -c 262144 \
--temp 1.0 --top-p 0.95 --top-k 20 --min-p 0.0 \
--presence-penalty 0.0 --repeat-penalty 1.0 --port 8000
Set context size and GPU offload to fit available memory. The example context setting is not a claim that these quantizations were evaluated at that length.
For image input, also download the projector and add --mmproj mmproj-Swift-Qwen3.8-Flash-Next-BF16.gguf to the server command:
hf download ukisai/Swift-1.5-Qwen3.8-Flash-Next-GSQ-RCO-GGUF mmproj-Swift-Qwen3.8-Flash-Next-BF16.gguf --local-dir .
This release reuses allocation search results rather than claiming a new RCO search on Swift. For IQ3_S the second expert pass also learns IQ4_NL codes (the other tiers keep IQ4_NL codes fixed and learn only their scales), and its calibration set adds German and French Wikipedia and GitHub code (sources disjoint from the reporting texts). The recipe summary identifies the selected variants and records the segmented-pass RNG limitation for IQ2_XS and Q2_0. The Swift V1MIX importance matrix and its provenance are included.
Per-tensor allocation dumps are provided in tensor-allocation, tied to the unsplit model identities in the release manifest.
GSQ and RCO were developed by the Deep Algorithms and Systems Lab at the Institute of Science and Technology Austria. This Swift adaptation is by UkisAI.
We acknowledge the Qwen team for the original model and ISTA-DASLab for the quantization methods and published allocations.
The Swift contribution is distributed under the Swift Open License v1.0. The original Qwen components retain the Qwen Community License 1.0. See NOTICE and the license texts for applicable terms.