This model is a fine-tuned version of Qwen/Qwen3-0.6B on the None dataset. It achieves the following results on the evaluation set: The following hyperparameters were used during training: - learningrate: 2e-05 - trainbatchsize: 4 - evalbatchsize: 8 - gradientaccumulationsteps: 16 - totaltrainbatchsize: 64 - lrschedulertype: cosine - lrschedulerwarmupsteps: 100 - numepochs: 3 - Transformers 5.17.0 - Pytorch 2.11.0+cu128 - Datasets 5.0.1 - Tokenizers 0.23.2
SAVRN Model Hub · Models by Task
Text Generation Models
1,582 open-weight text generation models in the SAVRN Model Hub, with Convergent Intelligence, Qwen and Francesca Padovani publishing the most.
1,582Models
732Publishers
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 | 25.2M | 1x MI300X, $1.85/hr |
| gpt2 | OpenAI community | 137M | mit | 18.8M | 1x MI300X, $1.85/hr |
| Qwen3-4B | Qwen | 4B | apache-2.0 | 11.4M | 1x MI300X, $1.85/hr |
| Qwen3-14B | Qwen | 14.8B | apache-2.0 | 9.4M | 1x MI300X, $1.85/hr |
| Qwen3-8B | Qwen | 8.2B | apache-2.0 | 8.1M | 1x MI300X, $1.85/hr |
| Qwen2.5-7B-Instruct | Qwen | 7.6B | apache-2.0 | 6.4M | 1x MI300X, $1.85/hr |
| gpt-oss-20b | OpenAI | 20.9B | apache-2.0 | 5.7M | 1x MI300X, $1.85/hr |
| opt-125m | AI at Meta | — | other | 5.6M | — |
| DeepSeek-V3.2 | DeepSeek | 685.4B | mit | 5.4M | 7x MI325X, $14.00/hr |
| Qwen2.5-0.5B-Instruct | Qwen | 494M | apache-2.0 | 5.3M | 1x MI300X, $1.85/hr |
Licenses
| License | Models | Commercial use |
|---|---|---|
| apache-2.0 | 734 | Yes |
| not stated | 345 | Not stated |
| other | 175 | Read the license |
| mit | 152 | Yes |
| cc-by-nc-4.0 | 42 | 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 1,582 Models, Page 15 of 27
GGUF build of openai/gpt-oss-20b for llama.cpp / Ollama / LM Studio. Converted by axteam with llama.cpp. The chat template (OpenAI harmony format) is embedded in the GGUF. Use a recent llama.cpp build (gpt-oss support, Aug-2025 or later). Reasoning effort can be set in the system prompt (Reasoning: low|medium|high). KV-cache type is a run-time choice in llama.cpp (not baked into the file): Apache-2.0, following the original openai/gpt-oss-20b. Weights, tokenizer, and chat template belong to OpenAI. This repo only re-packages them as GGUF. openai/gpt-oss-20b 의 GGUF 빌드 (llama.cpp·Ollama·LM Studio용). 변환: axteam (llama.cpp, chattemplate(OpenAI harmony)은 GGUF에 내장). - 파일…
The released GPT-2 large weights (774M parameters) fine-tuned into a chat model on with the loss on assistant replies only. Small, medium and large versions are compared in the instruction-tuning write-up. What is the capital of France? - best.safetensors: the weights at the lowest validation loss (the final step). - config.json: the training config. These weights use the GPT2 class from gpt2-repro, not Hugging Face Transformers, and a vocabulary of 50,263 tokens: GPT-2's plus six chat tokens. Clone gpt2-repro and install its dependencies with uv, then run the code below from the repository generatetext samples with temperature 1 and top-5 by default. The system message is optional. The…
The released GPT-2 medium weights (355M parameters) fine-tuned into a chat model on with the loss on assistant replies only. Small, medium and large versions are compared in the instruction-tuning write-up. What is the capital of France? - best.safetensors: the weights at the lowest validation loss (the final step). - config.json: the training config. These weights use the GPT2 class from gpt2-repro, not Hugging Face Transformers, and a vocabulary of 50,263 tokens: GPT-2's plus six chat tokens. Clone gpt2-repro and install its dependencies with uv, then run the code below from the repository generatetext samples with temperature 1 and top-5 by default. The system message is optional. The…
GPT-2 small (124M parameters) trained from random initialisation on the FineWeb-Edu sample-10BT sample, with the model and training loop written directly in PyTorch. Training details and plots: - gpt2step018890.safetensors: the final weights. - config.json: the training config. - checkpoints/step.pt: the last three resume checkpoints, with optimizer and RNG state, for continuing training with pretraining/train.py --resume. These weights use the GPT2 class from gpt2-repro, not Hugging Face Transformers. Clone gpt2-repro and install its dependencies with uv, then run the code below from the repository It was trained on educational web text in English only and can produce factually wrong or…
The released GPT-2 small weights (124M parameters) fine-tuned into a chat model on with the loss on assistant replies only. Small, medium and large versions are compared in the instruction-tuning write-up. What is the capital of France? - best.safetensors: the weights at the lowest validation loss (the final step). - config.json: the training config. These weights use the GPT2 class from gpt2-repro, not Hugging Face Transformers, and a vocabulary of 50,263 tokens: GPT-2's plus six chat tokens. Clone gpt2-repro and install its dependencies with uv, then run the code below from the repository generatetext samples with temperature 1 and top-5 by default. The system message is optional. The…
This is the lab4-yoru-AY-482206 checkpoint, trained from scratch on English C4. It was selected by the lowest development loss among nine experimental recipes. The independent seed repeat and reserved audit evaluation were still pending when this checkpoint was published. Reported scores are local evaluation proxies, not an official online-judge result. - Stock Hugging Face GPT2LMHeadModel: 12 layers, 12 attention heads, hidden width 768, context length 1,024, vocabulary 50,257, tied input/output embeddings. - 124,439,808 unique parameters, commonly described as GPT-2 small. The lab uses the historical 117M model-family label. - GPT-2 tokenizer; documents packed with EOS separators. No…
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
ZenDNN v6.1.0 - ZenTorch v2.13.0.0 - PyTorch v2.13.0.0 - LLM Compressor v0.13.0 - vLLM v0.29.0 This is a quantized version of granite-4.0-h-small created by AMD using LLM Compressor (compressed-tensors) for ZenDNN-optimized CPU inference. The model was quantized from granite-4.0-h-small using LLM Compressor with the GPTQ algorithm. This reduces the model weights from 60.0 GiB to 16.1 GiB on disk (~73% reduction). granite-4.0-h-small is a hybrid Mamba-MoE model: of its 40 layers, 4 are full-attention blocks and the other 36 are Mamba (linear-attention) blocks, and every layer carries a 72-expert MoE block (top-10 routing) alongside a shared MLP. Two details make this work. The model is…
ZenDNN v6.1.0 - ZenTorch v2.13.0.0 - PyTorch v2.13.0.0 - LLM Compressor v0.13.0 - vLLM v0.29.0 This is a quantized version of granite-4.0-h-small created by AMD using LLM Compressor (compressed-tensors) for ZenDNN-optimized CPU inference. The model was quantized from granite-4.0-h-small using LLM Compressor via the Round-to-Nearest (RTN) algorithm. This reduces the model weights from 60.0 GiB to 30.4 GiB on disk (~49% reduction). granite-4.0-h-small is a hybrid Mamba-MoE model: of its 40 layers, 4 are full-attention blocks and the other 36 are Mamba (linear-attention) blocks, and every layer carries a 72-expert MoE block (top-10 routing) alongside a shared MLP. The recipe only needs two…
ZenDNN v6.1.0 - ZenTorch v2.13.0.0 - PyTorch v2.13.0.0 - LLM Compressor v0.13.0 - vLLM v0.29.0 This is a quantized version of granite-4.0-h-tiny created by AMD using LLM Compressor (compressed-tensors) for ZenDNN-optimized CPU inference. The model was quantized from granite-4.0-h-tiny using LLM Compressor via the Round-to-Nearest (RTN) algorithm. This reduces the model weights from 12.9 GiB to 6.6 GiB on disk (~49% reduction). granite-4.0-h-tiny is a hybrid Mamba-MoE model: of its 40 layers, 4 are full-attention blocks and the other 36 are Mamba (linear-attention) blocks, and every layer carries a 64-expert MoE block alongside a shared MLP. The recipe only needs two ignore entries. lmhead…
以 Granite-4.2-30B-xVITA-zhTW（台灣繁中微調版）為來源，不是重新訓練，而是在權重上做一次正交化，把模型用來決定「要不要拒絕」的方向投影掉。 - 類型：因果語言模型，dense 29.3B、64 層 - 來源模型：xCloudinfo/Granite-4.2-30B-xVITA-zhTW（底座 ibm-granite/granite-4.2-30b） - 製作方式：單方向拒絕消除（weight orthogonalization，Arditi et al. 2024），非微調、非重訓 - 授權：Apache-2.0（沿用 Granite 底座授權） - 發布格式：GGUF（六階量化） 結論：去審查沒有顯著損及一般能力。 台灣口語行文風格完整保留（文體分還略高）；知識與程式各少答一題，落在小題庫雜訊範圍。可視為「開放版的 xVITA」。 本模型是云碩 xVITA 台灣繁中語音 RAG 套件組的生成＋工具層。完整迴圈（每一環都是自訓的繁中模型）： 語音聽寫（Whisper-v3-Turbo-xVITA）→ 語意檢索（xVITA-Embed）→ 重排序（xVITA-Rerank）→ 本模型：生成＋工具呼叫（看圖走 Gemma-4 xReadable OCR）→ 台灣腔語音合成（xVITA-TTS）。 答案踩在檢索到的台灣文件上重寫，台灣口語行文在 RAG 裡尤其加分（讀起來像台灣人寫的、不漂中國用語、不寫 AI 報告體）；整套可全地端私有化，適合政府與企業內部。生成層換成本開放版，還能處理被過度拒絕擋下的內部／資安問答。 IQ 階以校準語料計算…
An experimental MXFP4 quantization of ibm-granite/granite-4.2-3b, made with GPTQ calibration instead of the usual data-free rounding, and provided in two formats with identical weights. It's published as a data point, not as the recommended build. For most users, the Q4KM GGUF or GPTQ int4 build is the better choice. Format: OCP MX — E2M1 (FP4) weights, groupsize=32, one UE8M0 shared scale per group; weights only, activations stay 16-bit. Embeddings, norms and lmhead are unquantized. MXFP4 is usually produced data-free, by rounding weights straight to FP4 with no calibration data. llm-compressor ships an MXFP4A16 scheme, but every example we found pairs GPTQModifier with integer schemes…
A 4-bit GGUF of ibm-granite/granite-4.2-3b, quantized with an importance matrix built from a mixed prose and tool-calling calibration corpus. Of the three quantizations we made of this model, it's the closest to the original and the smallest. "Same top token" is llama.cpp's Same top p statistic: the share of positions where the quantized model's most likely next token matches the f16 model's. 1. Convert: converthftogguf.py (mainline llama.cpp), bf16 safetensors → f16 GGUF. Granite's scaling parameters (attentionscale, embeddingscale, residualscale, logitsscale) are supported natively. 2. Importance matrix: llama-imatrix over prose plus tool-calling conversations, rendered through Granite's…
A 4-bit weight-only GPTQ quantization of ibm-granite/granite-4.2-3b, saved in compressed-tensors format for vLLM. On an AMD Radeon Pro V620 (RDNA2), vLLM served it at 65 tok/s single-sequence and ~1,000 tok/s aggregate at 16 concurrent sequences. llm-compressor GPTQModifier: - int4, symmetric, weight-only (activations stay 16-bit) - groupsize=32, actorder="weight" - lmhead left unquantized 512 samples, maxseqlength=512 The tuning mattered. With the default W4A16 preset (groupsize=128), the same pipeline scored 0.153 mean KLD. Moving to groupsize=32 with actorder="weight" cut that by about 28%, for roughly 0.2 GB more disk. For comparison, an AWQ build (W4A16 asymmetric, groupsize=128) from…
This is a merged QLoRA derivative of Qwen/Qwen3-4B-Instruct-2507, published for Headroom, developed under the working name Greenlight. The original model identifier is retained for reproducibility. Training used 336 synthetic examples and 300 configured optimizer steps through Veri. Dataset SHA-256: ed827924d7ba4118fa15c5ef71fb211c557c3390d87384f21af4eb86ae59d051. Training job: 42d582a35d2f. Rank 8, 4-bit loading, learning rate 0.0001, sequence limit 2048, micro-batch 1, gradient accumulation 2. Prompt profile: reference-contract-v5. This model proposes bounded structured payment terms. Consult the repository’s matched frozen base/tuned evaluation before making any quality claim. The corpus…
GRPO with positive-advantage-gated sampled entropy regularization; exact equation and provenance are recorded in baseline metadata. Training explicitly renders enablethinking=False, recorded as thinkingmode: false in the scientific contract. The pinned tokenizer and chat template remain unchanged. Use the same explicit keyword at inference, particularly for Qwen3-1.7B, whose default template enables thinking. Every checkpoint in this repository requires explicit nonthinking provenance. This public repository retains every published optimizer-update adapter, including update zero (the initial, untrained adapter). Training is planned for 4 epochs on the same 2,048-question training split as…
GRPO with a detached correct-response likelihood-rank reward; exact equation and provenance are recorded in baseline metadata. Training explicitly renders enablethinking=False, recorded as thinkingmode: false in the scientific contract. The pinned tokenizer and chat template remain unchanged. Use the same explicit keyword at inference, particularly for Qwen3-1.7B, whose default template enables thinking. Every checkpoint in this repository requires explicit nonthinking provenance. This public repository retains every published optimizer-update adapter, including update zero (the initial, untrained adapter). Training is planned for 4 epochs on the same 2,048-question training split as the…
GRPO with positive-advantage-gated sampled entropy regularization; exact equation and provenance are recorded in baseline metadata. Training explicitly renders enablethinking=False, recorded as thinkingmode: false in the scientific contract. The pinned tokenizer and chat template remain unchanged. Use the same explicit keyword at inference, particularly for Qwen3-1.7B, whose default template enables thinking. Every checkpoint in this repository requires explicit nonthinking provenance. This public repository retains every published optimizer-update adapter, including update zero (the initial, untrained adapter). Training is planned for 4 epochs on the same 2,048-question training split as…
GRPO with a detached correct-response likelihood-rank reward; exact equation and provenance are recorded in baseline metadata. Training explicitly renders enablethinking=False, recorded as thinkingmode: false in the scientific contract. The pinned tokenizer and chat template remain unchanged. Use the same explicit keyword at inference, particularly for Qwen3-1.7B, whose default template enables thinking. Every checkpoint in this repository requires explicit nonthinking provenance. This public repository retains every published optimizer-update adapter, including update zero (the initial, untrained adapter). Training is planned for 4 epochs on the same 2,048-question training split as the…
Research artifact. This model has not been properly tested or evaluated and is not necessarily secure. Use at your own risk in production settings. This GrugMoE checkpoint combines Step92, Step12, and Step20 weights through two static merges. It has 26 layers, 256 routed experts per layer, and four selected routed experts per token. The exported weights contain 502 tensors totaling 134,157,778,944 bytes. First, start from Step92 and replace 32 routed experts per layer with Step20 experts selected using BFCL-Parity activation traces. Preserve Step92's remaining weights, router, tokenizer, chat template, and model metadata. Then blend shared-expert projections with 50% Step12 and attention…
Research artifact. This model has not been properly tested or evaluated and is not necessarily secure. Use at your own risk in production settings. This GrugMoE checkpoint combines Step92, Step12, and Step20 weights through two static merges. It has 26 layers, 256 routed experts per layer, and four selected routed experts per token. The exported weights contain 502 tensors totaling 134,157,778,944 bytes. First, start from Step92 and replace 32 routed experts per layer with Step20 experts selected using BFCL-Parity activation traces. Preserve Step92's remaining weights, router, tokenizer, chat template, and model metadata. Then blend each shared-expert projection with 75% Step12 and 25% of…
Research artifact. This model has not been properly tested or evaluated and is not necessarily secure. Use at your own risk in production settings. This GrugMoE checkpoint combines Step92, Step12, and Step20 weights through two static merges. It has 26 layers, 256 routed experts per layer, and four selected routed experts per token. The exported weights contain 502 tensors totaling 134,157,778,944 bytes. First, start from Step92 and replace 32 routed experts per layer with Step20 experts selected using BFCL-Parity activation traces. Preserve Step92's remaining weights, router, tokenizer, chat template, and model metadata. Then replace all shared-expert projection weights with Step12…
Weights for dedicated inference engine for Qwen3.8-Flash-Next on AMD Strix Halo (gfx1151). One model family, one GPU, every kernel written for this silicon. Questions and numbers from your own machine go to the Discord. Bugs go to the server repository's issues. These weights are in halogen's own.hgn format and will not load in transformers, vLLM, or llama.cpp. They exist to be mounted into the halogen-flash-server container. From 0.15 the checkpoint is qwen38-flash-next-v2.hgn, with its lookup table beside it. Download both (about 110 GiB). The model needs about 62 GiB resident plus the memory-mapped table. The server fetches exactly these when HALOGENDOWNLOAD names this repo on a new…
qwen38-flash-next-v2-abliterated.hgn is now a complete native-v2 conversion of the original revision 8336e613ea508b13c2159bd0f68965d97a606b95. It is not the r1/r2 writer transplant into peonist's checkpoint, and does not reuse stock HGN payloads or stock sidecars. The filename is retained so existing exact-file download links still resolve. immutable binary SHA-256 52e83980ba24913d44788c92f3c484619303f576b394cee8ae0760f0a23efdf3. 2 threads, trellis beam 128; policy bf16-hgn-v2-rtn-nf4-signed-rms-trellis-v1. This release is not the historical LDLQ-calibrated transplant. - Full HIP source verification and CPU-native format checks of the main, n-gram and vision outputs passed before…
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).
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).
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).
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).
static quants of https://huggingface.co/runelab-ai/Hati-Glint-SQL-9B-v1 For a convenient overview and download list, visit our model page for this model. weighted/imatrix quants seem not to be available (by me) at this time. If they do not show up a week or so after the static ones, I have probably not planned for them. Feel free to request them by opening a Community Discussion. If you are unsure how to use GGUF files, refer to one of TheBloke's READMEs for more details, including on how to concatenate multi-part files. (sorted by size, not necessarily quality. IQ-quants are often preferable over similar sized non-IQ quants) Here is a handy graph by ikawrakow comparing some lower-quality…
This model is a fine-tuned version of goldfish-models/hebhebr10mb. It has been trained using TRL. This model was trained with SFT.
4-bit, group-size-128 AutoRound quantization of Altworld/Hemmingway-1, a 27B-parameter fine-tune of Qwen3.8-27B. This repository is an independent quantization, not the original model release. See the original model card for training, intended use, and benchmark information. Quantization: AutoRound 0.15.1, bits=4, groupsize=128, packingformat=autoround:autoawq, seqlen=1024, nsamples=64, iters=50. Certain layers remain in higher precision, as specified in quantizationconfig.json. The included six-shard weight index and auxiliary modelextratensors.safetensors are needed together. Approximate download size: 18.65 GB (17.37 GiB). This is a quantization of an existing fine-tuned model, not a new…
A LoRA adapter that teaches Altworld/Hemmingway-1 (a 27B Qwen3.5 GDN model) to reason in the Tree-of-Thought "reasoning-steps → final justified answer" style of the codex5.5ToT dataset. - Altworld/Hemmingway-1 — a 27B-parameter Qwen35ForCausalLM (Qwen3.5 / GDN architecture): 48 linear-attention layers + 16 full-attention layers, 64 layers total, plus an MTP (multi-token-prediction) head. BF16 ≈ 54.6 GB. - AmelieSchreiber/codex5.5ToT — 3,890 high-quality reasoning records across 10 domains (math, physics, CS/algorithms, chemistry, …). Each record is a [system, user, assistant] chat where the assistant response is a reasoning tree: it explores branches, compares alternatives, and ends with a…
Proprietary on-device language model fine-tuned specifically for commercial document reconciliation, multi-currency invoice audits, purchase order verification, and 100% private financial workflows. This model runs 100% locally inside the Hijin desktop application. Customer invoices, bank statements, and spreadsheets never leave the user's computer.
Model · Text generation
hin-deva-100mb-after-ppt-Dp-100mb-packed-bfdiso-ckpt500_seed3407
This model is a fine-tuned version of francesca9805/hin-deva-100mb-ppt-Dp-100mb-packed-bfdisoseed3407. It has been trained using TRL. This model was trained with SFT.
Model · Text generation
hin-deva-100mb-after-ppt-mp-struct-core-100mb-ckpt500_seed3407
This model is a fine-tuned version of francesca9805/hin-deva-100mb-ppt-mp-struct-core-100mbseed3407. It has been trained using TRL. This model was trained with SFT.
Model · Text generation
hishab-qwen3.5-0.8b-pipe-signed-full-v12-trim38k-GGUF
This is an uncensored version of Aleph-Alpha/Kolibri-1 created with abliteration. This ablation was performed entirely using llama.cpp's GGUF model, without relying on Transformers, For the specific ablation method, please refer to cvector-generator. GGUFs come from Hob-forge/Kolibri-1-GGUF. The ablation did not modify all expert modules. Stock llama.cpp does not support the kolibri1 architecture yet. Apply kolibri1-llama.cpp.patch (included in this repo) to llama.cpp at upstream commit 836d571, then build: - Risk of Sensitive or Controversial Outputs: This model’s safety filtering has been significantly reduced, potentially generating sensitive, controversial, or inappropriate content.…
This is an uncensored version of TokenRhythm/NeoHorse-1-4B created with abliteration (see remove-refusals-with-transformers to know more about it). This is a crude, proof-of-concept implementation to remove refusals from an LLM model without using TransformerLens. Layers 5-17 are being ablated (0-based indexing). The MTP and Visual components were extracted from the original Qwen/Qwen3.5-4B and can provide excellent support. Use the latest llama.cpp, - Risk of Sensitive or Controversial Outputs: This model’s safety filtering has been significantly reduced, potentially generating sensitive, controversial, or inappropriate content. Users should exercise caution and rigorously review generated…
Introducing Hunter 4B by Lalakai Labs. A 4B-parameter conversational reasoning model, distributed as GGUF weights for fast local inference with llama.cpp, Ollama, LM Studio, and any other GGUF-compatible runtime. GGUF is a safe, mmap-friendly format — no pickle files here. Hunter reasons out loud: it works through problems step by step inside... traces before giving you the final answer — handy for math, code, and multi-step tasks. Despite its small footprint, Hunter-4B is built to rival the performance of models in the 10–15B parameter range on reasoning, coding, and cyber tasks. Hunter 4B is a strong fine-tune of the Qwen 3.5 4B, trained on ~180,000 high-density examples. No raw web…
Hush-Nano-Chat is an English, single-turn instruction-tuned version of Soulitude/Hush-Nano. It starts from the 22M-parameter pretrained model and uses supervised fine-tuning (SFT) on instruction–response pairs. Due to the model's limited parameters, its response can be inaccurate, incomplete, or inconsistent. Hush-Nano-Chat has the following features: The base model was pretrained on 8.5B tokens (8,554,042,292) drawn from the following subsets: Then it was fine-tuned on a mixture of the following datasets: unsloth/alpaca-cleaned, databricks/databricks-dolly-15k, and HuggingFaceH4/norobots. Only the assistant response and ending EOS token contribute to the training loss. Zero-shot normalized…
An Asymmetric Hybrid Q4KM (Q80 / Q6K / Q4KM) GGUF quantization of tencent/Hy-Memory-30B-A3B, calibrated with an importance matrix (imatrix) and tuned specifically for single 24 GB GPUs (RTX 3090 / RTX 4090) as well as zero-VRAM CPU sidecar deployments. Hy-Memory-30B-A3B (HYV3ForCausalLM / hyv3 architecture) is a sparse Mixture-of-Experts memory writer model with 30.66B total parameters and ~3.0B active parameters per token (39 MoE layers, 128 routed experts, top-8 routing + 1 shared expert, and 8:1 GQA). Because 94.5% of the model's parameters (28.97B) live inside the 128 sparse routed experts while only 5.5% (1.69B) make up the always-active backbone, uniform quantization leaves massive…
Hy3 with half of its routed experts removed by RAZOR, a training-free expert pruning method. Every MoE layer keeps 96 of its original 192 routed experts. No gradient updates or recovery training were applied: the retained weights are the base model's own weights. Pruning touches only the routed expert pool. Attention, shared experts, the embedding and the LM head are untouched, so the compute per token drops only by the share of expert FLOPs that the removed experts would have contributed. The other budget is Requires a Transformers build containing the native hyv3 implementation. Weights are bfloat16. RAZOR asks whether the surviving computation can replace an expert's function, rather…
Hy3 with a quarter of its routed experts removed by RAZOR, a training-free expert pruning method. Every MoE layer keeps 144 of its original 192 routed experts. No gradient updates or recovery training were applied: the retained weights are the base model's own weights. Pruning touches only the routed expert pool. Attention, shared experts, the embedding and the LM head are untouched, so the compute per token drops only by the share of expert FLOPs that the removed experts would have contributed. The other budget is Requires a Transformers build containing the native hyv3 implementation. Weights are bfloat16. RAZOR asks whether the surviving computation can replace an expert's function…
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 0.21.0
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).
Public artifacts for the iclr-debug branch of HazyResearch/MsTok. This repository currently establishes the experiment namespace and contains a training-budget audit. No new model weights or final benchmark claims are implied. Future selected checkpoints and evaluations will be organized by unique run ID and optimizer step. Training budgets must report raw token positions per component, dataset size, global batch, optimizer updates, and LR horizons. Generation reports must label supplied-level-zero versus unconditional sampling and record the reference model precision, entropy, scored-token counts, and per-seed results. See NCM-BUDGET-AUDIT.md for historical budget provenance and…
The models of Correct Answers, Invalid Traces: What Verifiable Grade-School Math Reveals About Chain-of-Thought Traces (Puduppully, Misra, Iyer, Kalwar, Palod and Kambhampati, 2026). Each is a 12-layer GPT-NeoX model (124M parameters) trained from scratch on iGSM for 100k steps at batch 512. They differ only in the trace that followed the problem during training. Code and the evaluation protocol are at https://github.com/ratishsp/igsm-trace-validity. The tokenizer is GPT-2's, with iGSM's special tokens 222, 223 and 224 for [PROB], [SOL] and [ANS]. evaluate.py in the code repository runs the paper's evaluation on any of these folders.
Inferno-Exp-1.0 is a 76.2M-parameter decoder-only base language model trained from scratch by AlgoDriveAI. It is a math-first model: its training data centers on arithmetic and math word problems, and it is geared toward the AxiomicLabs small-language-model benchmarks used in SLM challenges, ArithMark 3.0 and ArithMark 2.0 (results below). Future versions will build on this math foundation with improvements aimed at broader language understanding and logical reasoning. This is a base model (no instruction tuning). The model code ships in the repo, so load it with trustremotecode=True. All scores are zero-shot on the checkpoint in this repo. The ArithMark benchmarks were scored with…
This is the pinned Q4KM GGUF file for Minstrel's optional Advanced local language engine on Windows and Linux. The model weights are byte-identical to mradermacher/Qwen3.5-4B-GGUF at 1a5df2c0cba51dae8ac5888420360d8703707171; the GGUF source file SHA-256 is 51eafbc127f35598c8f1d2ec58b2520d6126c7d1195c4eca26832e63a2939d39. The signed Minstrel manifest and provenance file are included for slot detailed-local, artifact version 0.1.0. The reference upstream is Qwen/Qwen3.5-4B at 851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a. The GGUF source does not establish the exact Qwen revision it converted. The Apache-2.0 license file is copied from that reference snapshot. Minstrel currently keeps this choice…
This is a text-only, 4-bit affine/group-64 MLX derivative for Minstrel's optional Advanced local language engine. It is based on the pinned mlx-community/Qwen3.5-4B-4bit snapshot 0e7ffd5c629ef7719d4cbc04069232580bfa9d9c. The deriveqwen35text.py tool removed vision tensors and capped the packaged context at 8,192 tokens. The signed Minstrel manifest and provenance file are included. The manifest is for Minstrel's detailed-local slot, artifact version 0.1.0; it is not a general quality certification. The reference upstream is Qwen/Qwen3.5-4B at 851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a. The source conversion does not establish the exact Qwen revision it used. The Apache-2.0 license file is…
Model · Text generation
intern-s1-mini-assay-transfer-record-level-v27-bbb-martins-l2-step1575
Full fine-tuned checkpoint selected from the V27 BBB Martins L2 assay-transfer run. metric.json contains the selected checkpoint's validation metrics. The four sharded BF16 model weights, model configuration, and tokenizer files are included. The validation/assay-transfer-record-level-v27-bbb-martins-l2-intern/ folder contains the 76 model-only validation summaries and matching per-bucket CSVs from step 0 through step 1875. The selected step 1575 metrics are also recorded in metric.json. No Morgan baseline was run on validation; the model-versus-Morgan comparison used the test split (W&B run).
Herhangi bir HF modelini, bellek bütçesi altında, bit-birebir (kayıpsız) çalıştır. Model bitleri hiç atılmaz; katman katman seçmeli yüklenir, kullanılınca bırakılır. Tepe bellek = tek katman, model boyutu değil. Çıktı, tam-yükle referansla logit düzeyinde birebir aynı. Bit sıkıştırma değil (yüksek-entropi ağırlıkların kayıpsız sınırı ~1,3×'tir — fizik). Kayıpsız dağıtım/akış: yerleşik ayak-izi = model/N, N sınırsız. Ölçümler bulutta (Kaggle/Colab) gerçek koşudan. iris verify ile kendin doğrula. 1. Böl — model katman/tensör düzeyinde parçalanır. 2. Akıt — her parça gerektiğinde belleğe alınır, kullanılınca bırakılır (disk-offload yok). 3. Birleştir — aynı ağırlık + aynı sıra = tam modelle…
Questions
Which Text generation models are most downloaded?
By monthly downloads reported by the Hugging Face Hub: .