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).
SAVRN Model Hub · Models by Task
Text Generation Models
1,548 open-weight text generation models in the SAVRN Model Hub, with Convergent Intelligence, Qwen and Francesca Padovani publishing the most.
1,548Models
721Publishers
107K to 1.7TParameter range
8Licenses
SAVRN's Take
Text generation is where most of the attention goes, and where most of the money is spent. The models on this page run from tiny research checkpoints that fit on a laptop to frontier-class systems that need a full server of accelerators. The first question we ask about any of them is not how smart it is, but what it costs to keep it answering around the clock.
That cost comes down to three things: the size of the weights, the precision you run them at, and how much context each request carries. Moving from 16-bit to 4-bit cuts the memory a model needs by roughly four times, which is often the difference between one server and several. Mixture-of-experts models, such as the newer DeepSeek, Qwen and Nemotron releases, hold a large total parameter count but activate only a small share per token, so they answer faster than their size suggests while still needing the memory to hold every expert.
Licensing decides the rest. Apache 2.0 and MIT models can go straight into a commercial product. Llama, Gemma's earlier generations and models marked other each carry terms you need to read first, and a model with no stated license is research material until the publisher says otherwise. Our advice is to shortlist by license, size to the hardware you can actually run, then test the survivors on your own prompts. The table below is sorted by what people download, not by what we recommend.
Most Downloaded
| Model | Publisher | Parameters | License | Monthly downloads | Cheapest GPUs at 16-bit |
|---|---|---|---|---|---|
| Qwen3-0.6B | Qwen | 752M | apache-2.0 | 26.7M | 1x MI300X, $1.85/hr |
| gpt2 | OpenAI community | 137M | mit | 18.6M | 1x MI300X, $1.85/hr |
| Qwen3-4B | Qwen | 4B | apache-2.0 | 11.6M | 1x MI300X, $1.85/hr |
| Qwen3-14B | Qwen | 14.8B | apache-2.0 | 9.2M | 1x MI300X, $1.85/hr |
| Qwen3-8B | Qwen | 8.2B | apache-2.0 | 8.3M | 1x MI300X, $1.85/hr |
| Qwen2.5-7B-Instruct | Qwen | 7.6B | apache-2.0 | 6.6M | 1x MI300X, $1.85/hr |
| gpt-oss-20b | OpenAI | 20.9B | apache-2.0 | 5.8M | 1x MI300X, $1.85/hr |
| opt-125m | AI at Meta | — | other | 5.8M | — |
| Qwen2.5-0.5B-Instruct | Qwen | 494M | apache-2.0 | 5.5M | 1x MI300X, $1.85/hr |
| DeepSeek-V3.2 | DeepSeek | 685.4B | mit | 5.5M | 7x MI325X, $14.00/hr |
Licenses
| License | Models | Commercial use |
|---|---|---|
| apache-2.0 | 714 | Yes |
| not stated | 341 | Not stated |
| other | 168 | Read the license |
| mit | 151 | Yes |
| cc-by-nc-4.0 | 41 | Not without separate permission |
| llama3.2 | 33 | Read the license |
| gemma | 22 | Yes, with conditions |
| llama3.1 | 21 | Yes, with conditions |
Who Publishes Them
| Publisher | Models |
|---|---|
| Convergent Intelligence | 54 |
| Qwen | 42 |
| Francesca Padovani | 37 |
| Yanggangu | 28 |
| Shreyansh singh | 25 |
| Kumar | 23 |
All 1,548 Models, Page 19 of 26
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).
Model · Text generation
nor-latn-100mb-after-ppt-Dp-100mb-packed-bfdiso-ckpt500_seed3407
This model is a fine-tuned version of francesca9805/nor-latn-100mb-ppt-Dp-100mb-packed-bfdisoseed3407. It has been trained using TRL. This model was trained with SFT.
Model · Text generation
nor-latn-100mb-after-ppt-mp-struct-100mb-ckpt500_seed3407
This model is a fine-tuned version of francesca9805/nor-latn-100mb-ppt-mp-struct-100mbseed3407. It has been trained using TRL. This model was trained with SFT.
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). - PEFT 0.21.1
GLM-5.3 uses the same base model as GLM-5.2 — every gain comes from post-training. Compared with GLM-5.2, it is much better at complex coding and long-horizon tasks: GLM-5.3 supports deployment with the following frameworks. Feel free to try them out: - SGLang — see cookbook - vLLM — see recipes - TokenSpeed — see here - Transformers — see transformers docs - KTransformers — see tutorial - Unsloth — see guide - For deployment on the Ascend NPU platform, inference frameworks such as vLLM-Ascend, xLLM and SGLang are supported — see here. - GLM-5.3 supports controlling the thinking budget through the reasoningeffort parameter, which accepts three levels: low, high, and max. It defaults to max…
Language Intelligence layer of the NULLXES Intelligence Stack. SHINRA Our llm. Release line 1. NULLXES SHINRA-4B-BASE — pretrain 2. NULLXES SHINRA-4B-INSTRUCT — instruction tuning ← this model 3. NULLXES SHINRA-4B-INSTRUCT (aligned) — DPO / preference optimization Parameter budget Architecture source: configs/shinra4b.yaml at 903a639e03bc. The total also matches the published safetensors metadata. Custom SentencePiece Unigram, trained in-house on a web + wiki + code mix. Special tokens trustremotecode=True is required: SHINRA ships a custom ShinraConfig and modeling code, not a reused architecture class. The HF inference widget cannot load custom code, hence inference: false. Data pipeline…
NULLXES SHINRA-4B-INSTRUCT is the Language Intelligence Layer of the NULLXES system. SHINRA is responsible for multilingual understanding, coding intelligence, instruction following, structured outputs, and agent preparation. This checkpoint is the instruction-tuned (and optionally DPO-aligned) 4B-class dense decoder. Proprietary ShinraForCausalLM (not a Llama / Mistral / Qwen / GPT-NeoX wrapper). RMSNorm → GQA+RoPE → residual → RMSNorm → SwiGLU → residual then final RMSNorm and tied LM head. Special tokens:. Generation stop is. Document stop is. Three stages. Pretrain → NULLXES SHINRA-4B-BASE SHINRAPRETRAINV1: 40% FineWeb-Edu, 20% code (python-edu + licensed Stack), 15% math/science…
GGUF quantized version of nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16. 1. Upstream llama.cpp: Model probably works. 2. Forked version of llama.cpp: RPC with split mode tensor, MTP, and model works - Generate Korean-focused imatrix calibration dataset (Total 512 samples, Chat template applied for messages) - Create imatrix using llama-imatrix with --parse-special flag Unfortunately, I don't have PPL/KLD for quant, due to out of compute resource and target budget. 2x DGX Sparks, RPC, -sm tensor with MTP enabled (-ub 2048 -b 2048 -c 262144)
An MLX 4-bit conversion of NVIDIA's for Apple silicon. This is a community conversion. It is not made, reviewed, or endorsed by NVIDIA. For the model's design, training data, benchmarks, and intended use, see Dory 1.0 4B is a looped (latent-recurrent) normalized-GPT model trained by NVIDIA. Its layers form three blocks: input, recurrent, and output. The shared recurrent block runs several times per token, and the input block's output is added back at the start of each loop. More loops give more compute per token and better accuracy from the same weights. It mixes 1,024-token sliding-window attention with full attention, supports up to 256K tokens of context, and reasons in a block before it…
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). - PEFT 0.19.1
The model weights of the LoRA / QLoRA runs done on the L40S server during the Warsaw Model Trainers Hackathon (25-26.09.2026): PEFT adapters, Trainer checkpoints (with optimizer state), lora.gguf adapters for llama.cpp and trainingargs.bin. These are adapters, not full models. Every run uses the base google/gemma-4-12B-it-qat-q40-unquantized (the local path in adapterconfig.json points to a server copy of it). Layout: l40s-results/ /..., the same paths as oldwb/l40s-results/ in the GitHub repo Endote/WarsawModelTrainersHackathon, which holds the logs, metrics, data statistics and write-ups (README.md, EXPERIMENTS.md, BENCHMARKSRESEARCH.md). Restore into a checkout of that repo: python…
This model is a fine-tuned version of allenai/OLMo-2-1124-7B-Instruct. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of allenai/OLMo-2-1124-7B-Instruct on the Neelectric/wildguardmixLlama-3.1-8B-Instruct4096toks dataset. It has been trained using TRL. This model was trained with SFT.
基于 MikeLee000/olmo3-190m-zh-full 完成一轮持续预训练（Continued Pretraining，CPT）的课程实验模型，约 187M 参数。 不是 SFT 或指令对话模型。 使用普通文本作为提示进行续写，不使用 applychattemplate，不应将其作为可靠的问答助手。 - 架构：Olmo3ForCausalLM，12 层，hidden size 768，intermediate size 3072，词表 48,000。 - 数据：cmz1024/llm101-olmo3-zh-demo-data 仓库中的 continue/tokenizedv31.bin。 - 全参数持续预训练；不使用 LoRA；序列长度 2,048；BF16。 - 1 epoch，共 11,918 个 optimizer steps。 - Batch size 16，gradient accumulation 4，有效 batch size 64。 - Learning rate 2e-4，warmup ratio 0.1，cosine scheduler。 - Transformers 5.9.0，PyTorch 2.5.1。 - 最终 validation loss：1.9467759132385254。这是特定验证集上的语言模型损失，不是准确率或通用问答质量评分。 发布文件来自 /checkpoints/olmo3zhcontinue/final，已核对与 checkpoint-11918/model.safetensors 的 SHA-256 一致：…
Standalone merged BF16 model from epoch 3.970178926441352, step 1500. Dataset: CompassioninMachineLearning/urban12738cleaned at ef7c0e742df63ea319e35d02d9f6ba63d6e7c68d. Training: 10,072 distinct documents plus 2,000 repeat exposures per epoch; 200 disjoint validation documents. Merged with Unsloth's native savepretrainedmerged(savemethod="merged16bit"). Weights are validated BF16 and packaged losslessly into eight safetensors shards. No adapter is required to load this model. See runmanifest.json for base revision, document selection hashes, training parameters and export validation. Training does not establish an improvement in compassion; evaluate that separately.
basemodel: ibm-granite/granite-4.1-3b libraryname: peft pipelinetag: text-generation - lora - qlora - peft - oracle - oracle19c - nl2sql - text-to-sql - sql-generation - granite A QLoRA adapter fine-tuned to convert natural-language questions into Oracle 19c SQL queries. This repository contains a LoRA adapter, not a full standalone model. - Convert natural-language questions into Oracle 19c SQL - Use with a schema-retrieval or RAG layer for better table and column selection - 2,800 Oracle-focused NL→SQL examples - 400 general-SQL replay examples - This adapter is not a database-security system. - Review every generated SQL query before running it. - Use read-only database access for…
We maintain our previous releases publicly as a transparent engineering record of continuous optimization. Below is the exact evolutionary roadmap of our MiniPlus architectures: The APEX-I-MiniPlus quantization architecture powering this model was subjected to an extensive independent evaluation by Japanese AI researcher and evaluator zephel01 (CoolZero) on an NVIDIA RTX 5090 (32GB) workstation running llama.cpp CUDA b11027 with FlashAttention (-fa on -ctk q80 -ctv q80 -ngl 99). The evaluation tested the APEX-I hybrid MoE engine across 348 unseeded trials on SWE-bench style multi-file Python bug-fixing tasks with hidden pytest suites (llmbench): - L6 Multi-File Code Generation (60 tasks)…
A fine mix-quantified and 50REAP version for Ornith-1.5 that is specially designed for 12GB cards. The model runs perfectly inside a llama.cpp. And maximum context of 256K can be reached with params below on 12G card. Suggested params: -c 262144 -ngl 99 -ctk q51 -ctv q50 -fa on --jinja -np 1 --temp 1 --top-k 20 --top-p 0.95 --min-p 0.10 Test runs on RTX5070ti laptop GPU(12G) with 90token/s of max speed. Real agent coding have an output approxly 50t/s tested on Kimicode.
Vietnamese spelling correction model: a BARTpho / MBart seq2seq model (~400M parameters), quantized to INT8 for inference on CPU. Two formats ship in this repository: CTranslate2 (ct2/) and ONNX Runtime (onnx/). This repository holds the model weights used by the Python package orthomni-codeco on PyPI. You normally don't download these files by hand: the package fetches the files of the backend you use on first run. Requires Python >= 3.9. PyTorch is not needed: the default runtime dependencies are only ctranslate2, numpy, sentencepiece and huggingface-hub. Create the model once and reuse it. Later runs load from the local cache and work offline. An async variant is also available: To use…
This model is a fine-tuned version of Qwen/Qwen2.5-0.5B-Instruct on an unknown dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 2e-05 - trainbatchsize: 2 - evalbatchsize: 8 - lrschedulertype: linear - trainingsteps: 2000 - Transformers 5.20.0 - Pytorch 2.14.1+cu130 - Tokenizers 0.23.3
The on-device storyteller for Outward, an offline walking adventure. You go outside, find something real ("rainwater running down a drain"), and this model writes the next 60–120 words of the story around it. It runs entirely on the phone, in the browser, with llama.cpp (wllama). Modified by the Outward project. This is a Model Derivative of google/gemma-3-270m-it: fully fine-tuned on 3,186 stories written by a LoRA fine-tuned Qwen3.5-4B teacher (trained on Tinker), then converted to GGUF and quantized to Q40. The eval set has 50 held-out prompts × 3 samples, temperature 0.7, scored with the automatic rubric in the Outward repo: uses the discovery, 60–120 words, safe and outdoors (risky…
Inference weights and tokenizer for the historical 1-data experiment. Part of the OVD collection.
Inference weights and tokenizer for the historical 1-data experiment. Part of the OVD collection.
OzyAssist Voice es un modelo de inteligencia artificial de ultra-baja latencia (30+ tokens/segundo, tiempo de respuesta <330 ms), destilado y afinado específicamente para asistentes de voz interactivos, cowork en vivo, livestreaming y control de Windows. Diseñado bajo la arquitectura Zero-Docker de OzyAssist, este modelo corre localmente consumiendo apenas ~1.37 GB de VRAM, lo que permite operarlo en cualquier laptop o GPU mientras juegas o transmites en directo. Pruebas empíricas ejecutadas en el mismo entorno de hardware (NVIDIA Quadro RTX 4000 Max-Q): Descarga el modelo GGUF y ejecútalo con aceleración GPU: Crea un archivo Modelfile: Luego construye y corre el modelo: En el proyecto…
Peta1 is an autonomous software engineering agent checkpoint based on google/gemma-4-E4B-it (4.3B active parameters), post-trained using a targeted three-stage curriculum: 1. Tree-of-Reasoning (ToR) SFT: Supervised deliberation in Gemma 4's native latent thinking tokens ( thought) enforcing hypothesis formulation, environment simulation, and branch pruning. 2. Capability Frontier Discovery: Screening of candidate tasks (G = 4, T = 0.7) to isolate the 18% informative frontier band (0 < p < 1.0), preventing policy gradient collapse from trivial (36%) and impossible (46%) tasks. 3. Verifiable Outcome GRPO: Policy gradient optimization with verifiable rewards (Rtarget + Rformat + Rtor - lambda…
petal-w1 turns raw dictation into the text you meant to type. It is the on-device cleanup model in Petal, a free macOS dictation app. It runs with MLX on Apple silicon. It is a full fine-tune of Qwen3.5-0.8B, and it does one job: - Removes fillers, stutters, false starts, and verbal tics ("you know what I mean?"). - Keeps only the final version of a self-correction ("Friday, no wait, Thursday" becomes "Thursday"), also when speech recognition mishears the correction words. - States a point once when the speaker repeats it in different words, and keeps every detail. - Condenses rambles into clear sentences, and keeps every fact, name, number, and request. - Keeps questions as questions and…
petal-w1 turns raw dictation into the text you meant to type. It is the on-device cleanup model in Petal, a free macOS dictation app. It runs with MLX on Apple silicon. It is a full fine-tune of Qwen3.5-0.8B, and it does one job: - Removes fillers, stutters, false starts, and verbal tics ("you know what I mean?"). - Keeps only the final version of a self-correction ("Friday, no wait, Thursday" becomes "Thursday"), also when speech recognition mishears the correction words. - States a point once when the speaker repeats it in different words, and keeps every detail. - Condenses rambles into clear sentences, and keeps every fact, name, number, and request. - Keeps questions as questions and…
A specialized 3.8B biomedical & clinical reasoning foundation model built on Microsoft's Phi-4-mini-instruct, formatted for standard Hugging Face transformers and PyTorch. 1. Stage 1 (STEM Foundation): 116,000 instruction pairs across NCERT Classes 6–12 (Physics, Chemistry, Biology) eliminating foundational science hallucinations. 2. Stage 2 (PubMed 2026 Evidence): 12 recent 2026 clinical update archives from NCBI FTP covering survival outcomes (OS, PFS, HR), targeted therapeutics, and clinical trial endpoints. 3. Stage 3 (Comprehensive Internal Medicine): Balanced multi-specialty clinical curriculum (cardiology, nephrology, endocrinology, pulmonology) with an active oncology replay buffer.…
A specialized 3.8B biomedical & clinical reasoning model built on Microsoft's Phi-4-mini-instruct, optimized natively for Apple Silicon Metal acceleration via Apple MLX. The model underwent a 3-stage transfer learning curriculum: 1. Stage 1 (STEM Foundation): 116,000 instruction pairs across NCERT Classes 6–12 (Physics, Chemistry, Biology) eliminating foundational science hallucinations. 2. Stage 2 (PubMed 2026 Evidence): 12 recent 2026 clinical update archives from NCBI FTP covering survival outcomes (OS, PFS, HR), targeted therapeutics, and clinical trial endpoints. 3. Stage 3 (Comprehensive Internal Medicine): Balanced multi-specialty clinical curriculum (cardiology, nephrology…
Rosa is Patriot Memory's 253M-parameter edge assistant for English and Traditional Chinese — built to fit the edge devices you actually ship, with a vocabulary trained natively on Traditional Chinese. It provides accurate information regarding: If you run into multi-GPU tensor device mismatch errors: RuntimeError: Expected all tensors to be on the same device... Run the script with CUDAVISIBLEDEVICES=0 to isolate execution to GPU 0. Good: introducing herself in both languages; answering arithmetic word problems with visible English reasoning steps; clean Traditional Chinese — glyphs, vocabulary and register; running fully offline on edge hardware. Not: her reasoning traces are formatted…
Bro, imagine an AI that can see, hear, and read a website while hunting bugs at the same time. Yeah, that's PolyglotBugHunter-X. Try it live — one click, no install, no server, runs in your browser. A multimodal agent for automated web security testing. It reads pages (text), looks at rendered output (image), and listens to uploads (audio) at the same time, then reports what it found with CVSS v3.1 scores and evidence you can re-check. It ships 43 detection payloads across 8 bug classes (24 of them polyglot — single strings that are simultaneously plausible HTML, JS, SQL and shell), a payload-routing engine that guesses the right vector from the parameter name, and a scoring profile. This…
Model · Text generation
ppt-pythia-160m-uniform250-previous_mse_delta_shuffle1-seed324-stage2
This model is a fine-tuned version of qing-yao/ppt-pythia-160m-uniform250-previousmse-seed324-stage1 on the None dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 0.001 - trainbatchsize: 16 - evalbatchsize: 16 - gradientaccumulationsteps: 2 - totaltrainbatchsize: 32 - lrschedulertype: cosinewithminlr - lrschedulerwarmupsteps: 500 - trainingsteps: 10000 - Transformers 5.4.0 - Pytorch 2.8.0+cu128 - Datasets 3.2.0 - Tokenizers 0.22.1
Model · Text generation
ppt-pythia-160m-uniform250-previous_mse_delta_shuffle2-seed324-stage2
This model is a fine-tuned version of qing-yao/ppt-pythia-160m-uniform250-previousmse-seed324-stage1 on the None dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 0.001 - trainbatchsize: 16 - evalbatchsize: 16 - gradientaccumulationsteps: 2 - totaltrainbatchsize: 32 - lrschedulertype: cosinewithminlr - lrschedulerwarmupsteps: 500 - trainingsteps: 10000 - Transformers 5.4.0 - Pytorch 2.8.0+cu128 - Datasets 3.2.0 - Tokenizers 0.22.1
This model is a fine-tuned version of EleutherAI/pythia-160m on an unknown dataset. The following hyperparameters were used during training: - learningrate: 0.001 - trainbatchsize: 16 - evalbatchsize: 16 - gradientaccumulationsteps: 2 - totaltrainbatchsize: 32 - lrschedulertype: cosinewithminlr - lrschedulerwarmupsteps: 13 - trainingsteps: 250 - Transformers 5.4.0 - Pytorch 2.8.0+cu128 - Datasets 3.2.0 - Tokenizers 0.22.1
This model is a fine-tuned version of EleutherAI/pythia-160m on an unknown dataset. The following hyperparameters were used during training: - learningrate: 0.001 - trainbatchsize: 16 - evalbatchsize: 16 - gradientaccumulationsteps: 2 - totaltrainbatchsize: 32 - lrschedulertype: cosinewithminlr - lrschedulerwarmupsteps: 13 - trainingsteps: 250 - Transformers 5.4.0 - Pytorch 2.8.0+cu128 - Datasets 3.2.0 - Tokenizers 0.22.1
This model is a fine-tuned version of EleutherAI/pythia-160m on an unknown dataset. The following hyperparameters were used during training: - learningrate: 0.001 - trainbatchsize: 16 - evalbatchsize: 16 - gradientaccumulationsteps: 2 - totaltrainbatchsize: 32 - lrschedulertype: cosinewithminlr - lrschedulerwarmupsteps: 13 - trainingsteps: 250 - Transformers 5.4.0 - Pytorch 2.8.0+cu128 - Datasets 3.2.0 - Tokenizers 0.22.1
This model is a fine-tuned version of EleutherAI/pythia-1b. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of EleutherAI/pythia-1b. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of EleutherAI/pythia-1b. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of EleutherAI/pythia-1b. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of EleutherAI/pythia-1b. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of EleutherAI/pythia-1b. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of None. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of None. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of None. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of None. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of EleutherAI/pythia-1b. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of EleutherAI/pythia-1b. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of None. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of None. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of None. It has been trained using TRL. This model was trained with SFT.
This model is a fine-tuned version of None. It has been trained using TRL. This model was trained with SFT.
Configuration and automation for local LLM and image-generation inference on a dual-NVIDIA GPU host. The repository includes llama.cpp model presets, a llama-swap gateway configuration, and Docker-based build/run scripts for llama.cpp, stable-diffusion.cpp, and Unsloth Studio. - CUDA 12.9.1 llama.cpp CUDA builds target both GPUs with: The host keeps CUDA 13.4. The llama.cpp build uses CUDA 12.9 inside an Ubuntu 24.04 container because CUDA 12.9 is incompatible with host Ubuntu 26 headers. The repository has two preset layers: - preset.ini is the multi-model llama.cpp router preset used by the optional standalone service in scripts/llama-server.sh. - llama-swap-presets/.ini contains one…
The Prism Roleplay 1.5 Small recipe on PrismML's 1-bit Bonsai-8B: a standalone model, faster and smaller than the 4-bit original, with a small quality gap A standalone model, not an adapter: load it directly with mlx-lm. It is Prism Roleplay 1.5 Small's training recipe (same data, hyperparameters and system prompt) applied to prism-ml/Bonsai-8B-mlx-1bit, a 1-bit (g128) build of Qwen3-8B. How the LoRA was merged into a 1-bit model. A plain merge re-quantizes every layer to 1 bit, and the small LoRA update disappears in the rounding (we measured it: merged held-out loss 3.415 = untuned base 3.405). So the 16 layers the LoRA touched (all their attention and MLP projections) are merged at fp16…
PrismaCoder-3B is a Qwen2.5-Coder-3B-Instruct fine-tune specialized for generating and editing Prisma ORM schemas and client code from natural-language instructions. It was trained with a full SFT → DPO → GRPO pipeline, using a verifier (gate) as both the data-quality filter and the reinforcement-learning reward signal. stage, LoRA weights merged into the base model between stages iterative self-play RL round (attempted; see Limitations) weights for release All stages used seed 20260830 and ran on a single Kaggle T4 GPU (16GB) with gradient checkpointing and paged 8-bit AdamW. verifier (LightValidator, "V0" in the project's internal gate numbering): it checks that generated Prisma…
Questions
Which Text generation models are most downloaded?
By monthly downloads reported by the Hugging Face Hub: .