bradshaw-ai/Swift-1.5-27B-5.65bpw-GSQ-RCO-Marlin-MTP

text-generation model by bradshaw-ai

From the publisher

Model card & documentation

Source preview

Read the publisher’s intended use, setup instructions, evaluations and limitations. The original model card is the source of truth.

I created Swift-1.5-27B-5.65bpw-GSQ-RCO-Marlin-MTP (based on UkisAI/Swift-1.5-Qwen3.8-27B) for dual 24 GiB GPU workstations—specifically two RTX 3090s running Tensor Parallelism (TP=2) over PCIe Gen4 x8/x8 without NVLink. When serving a 27B hybrid Gated DeltaNet (GDN) + Softmax GQA model on consumer multi-GPU hardware, I wanted to solve two practical bottlenecks: To address both, this checkpoint combines Riemannian Constrained Optimization (RCO) to allocate vLLM Marlin kernel precisions across the 64-layer backbone, Gumbel-Softmax Quantization (GSQ) + AutoRound to minimize quantization distortion, a 5-layer W8 exit corridor (L59–L63), and an in-situ self-distilled, 32k English-pruned W4G32/W8G128 MTP speculative drafter designed to run with the syv-ai/HyperQwen vLLM patch suite. Measured on 2× NVIDIA RTX 3090 (47.11 GiB usable VRAM, sm86, PCIe Gen4 x8/x8 without NVLink) running vLLM (v0.30.0 + syv-ai/HyperQwen patches) with --speculative-config '{"method":"mtp","numspeculativetokens":4,"draftsamplemethod":"probabilistic"}': The 64-layer hybrid backbone interleaves 48 Gated DeltaNet (GDN) linear-attention layers with 16 Full Softmax GQA layers (l ∈ {3, 7, 11, ...,…

bradshaw-ai/Swift-1.5-27B-5.65bpw-GSQ-RCO-Marlin-MTP on Hugging Face A short preview, not the full document.

Read the full model card ↗ · Preview checked 2026-11-03T07:18:27.602Z

Inside the original model card — Document outline

Headings are captured from the source. Links open the publisher’s document, not a locally hosted copy.

Documentation belongs to its respective authors. Reported project/model license: other. A listing is not a grant of reuse or training rights. Confirm the document’s own terms at the source.

Model overview

bradshaw-ai/Swift-1.5-27B-5.65bpw-GSQ-RCO-Marlin-MTP is a text-generation model repository published by bradshaw-ai on Hugging Face. A library was not reported.

This page summarizes Hub metadata. For intended use, training data, evaluation results and limitations, consult the original model card.

Model facts

Task: text-generation
Library: Not reported
Recent downloads (30 days): 0
Cumulative likes: 0
Hugging Face trending score: 0
Reported safetensors parameters: Not reported
Architecture: Not reported
License: other
Access: Not gated by Hugging Face
Created: 2026-11-03T06:18:54.000Z
Last modified: 2026-11-03T06:31:13.000Z

Compatibility and lineage

Language coverage was not reported.

Reported base models: UkisAI/Swift-1.5-Qwen3.8-27B

Parameter count is not a RAM/VRAM requirement. Check precision, quantization, context length and runtime compatibility in the model card; no hardware or API-cost claim is inferred here.

Use and evaluate

Check license terms and access requirements first. Review the model card’s documented loading instructions, evaluations and safety limitations. Benchmark scores are not imported by this directory, and popularity is not an accuracy ranking.

No model weights are downloaded or executed by AltAPIs. Never enable remote model code without reviewing it.

Model card and usage instructions

Model files and configuration

Community discussion

Source and freshness

Source: Hugging Face Hub. Metadata observed 2026-11-03T06:52:20.270Z. Daily imports are snapshots, not real-time monitoring.

Popularity and source listings do not establish security, suitability, licensing rights or benchmark performance.

Open original source

Related models

openai-community/gpt2

deepseek-ai/DeepSeek-V3.2