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 25 of 26
tanpo-deals (Deal Cracker) is a compact B2B deal-closing and distributor/channel specialist for practical commercial work—not generic chatbot chatter. This repository contains the merged Transformers fine-tune, ready to load with transformers or Unsloth. Official DarkLab evaluation on the same 20-task deals rubric and decoding setup. tanpo-deals beats the base by +7.3 percentage points. BEATSBASE: YES Automated rubric results are directional; human judgment remains important for consequential commercial decisions. - GGUF for LM Studio / llama.cpp: d4rkninja/tanpo-deals-GGUF — recommend Q4KM Focused on B2B sales, deal closing, wholesale, distributors, channel partners, negotiation, and…
Ready-to-run GGUF quantizations of tanpo-deals, a compact B2B deal-closing and distributor/channel specialist. Open a.gguf in LM Studio, llama.cpp, Ollama, or another GGUF-compatible runtime—no LoRA wiring required. Official DarkLab evaluation on the same 20-task deals rubric and decoding setup. The tanpo-deals fine-tune beats the base by +7.3 percentage points. BEATSBASE: YES Discovery and qualification; objection handling and closing; proposal and ROI; distributor recruiting; channel-partner enablement; wholesale pricing and terms; negotiation tactics; pipeline and forecast hygiene; outreach sequences; and identity/safety guardrails. Automated rubric results are directional; verify…
A compact customer-retention specialist (~1.2B) for onboarding, activation, churn-save, customer-success playbooks, health-score follow-up, renewals, expansion, escalation, identity guardrails, and winback workflows — built for local and inexpensive deployment. Original upstream: LiquidAI/LFM2.5-1.2B-Instruct (~1.17B parameters, 32,768-token context, designed for edge/on-device deployment). Fine-tuning: Unsloth-compatible loading of that checkpoint via hub id unsloth/LFM2.5-1.2B-Instruct (LoRA / PEFT). This repository hosts the merged Transformers weights (LoRA merged into the base). Tanpo is a family of compact domain-specialized business models for local / edge / inexpensive deployment.…
GGUF quantizations of d4rkninja/tanpo-retention for local runtimes (llama.cpp, LM Studio, and other GGUF-compatible tools). Tanpo is a family of compact domain-specialized business models for local / edge / inexpensive deployment. Different specialists cover different workflows. Original upstream: LiquidAI/LFM2.5-1.2B-Instruct (~1.17B parameters, 32,768-token context, designed for edge/on-device deployment). Fine-tuning: Unsloth-compatible loading of that checkpoint via hub id unsloth/LFM2.5-1.2B-Instruct (LoRA / PEFT). A compact customer-retention specialist (~1.2B) for onboarding, activation, churn-save, customer-success playbooks, health-score follow-up, renewals, expansion, escalation…
PEFT LoRA adapter that specializes LiquidAI/LFM2.5-1.2B-Instruct for retention workflows (trained via Unsloth hub id unsloth/LFM2.5-1.2B-Instruct). Tanpo is a family of compact domain-specialized business models for local / edge / inexpensive deployment. Different specialists cover different workflows. Original upstream: LiquidAI/LFM2.5-1.2B-Instruct (~1.17B parameters, 32,768-token context, designed for edge/on-device deployment). Fine-tuning: Unsloth-compatible loading of that checkpoint via hub id unsloth/LFM2.5-1.2B-Instruct (LoRA / PEFT). Full eval, examples, and responsible-use notes: d4rkninja/tanpo-retention. - Onboarding, activation, churn-save, renewal, expansion, escalation, and…
GPT-Neo storyteller (nominal 1M; actual 3,745,984 parameters including the 50,257-token embedding) trained on TeenyTinyStories (TTS): 2,491,490 stories (498.5M tokens with EOS) generated by roneneldan/TinyStories-Instruct-33M. batch 1280 x 512, constant LR 5e-4, weight decay 0.1, betas (0.9, 0.95), clip 1.0. - TS-valid capped to stories of at most 400 tokens (20,711 stories): 2.369 / ppl 10.69 Research on small-scale synthetic pretraining data. Generates simple children's stories; sample at temperature 0.8-1.0. See the TeenyTinyStories paper for the corpus recipe and judge evaluations. Apache-2.0.
GPT-Neo storyteller (nominal 33M; actual 68,514,048 parameters including the 50,257-token embedding) trained on TeenyTinyStories (ableian/TeenyTinyStories) batch 1280 x 512, constant LR 5e-4, weight decay 0.1, betas (0.9, 0.95), clip 1.0. - TS-valid capped to stories of at most 400 tokens (20,711 stories): 1.539 / ppl 4.66 Research on small-scale synthetic pretraining data. Generates simple children's stories; sample at temperature 0.8-1.0. See the TeenyTinyStories paper for the corpus recipe and judge evaluations. Apache-2.0.
This model is a fine-tuned version of /helios-storage/helios4-data/cuong/Terminal-complete8k on the nemotroncompletebottom8012k dataset. The following hyperparameters were used during training: - learningrate: 1e-05 - trainbatchsize: 1 - evalbatchsize: 8 - distributedtype: multi-GPU - numdevices: 4 - totaltrainbatchsize: 4 - totalevalbatchsize: 32 - lrschedulertype: cosine - numepochs: 2.0 - Transformers 5.6.0 - Pytorch 2.11.0+cu130 - Datasets 4.0.0 - Tokenizers 0.22.2
Refusal-ablated Ternary Bonsai 2 27B, edited directly on the ternary lattice and shipped in PrismML's native PTQ10 pack. - 5,946,648,928 bytes — byte-for-byte the size of the official PTQ10 release - No requantization, no BF16 bake, no runtime steering, no control vector at serve time - Requires the PrismML llama.cpp fork Differs from PrismML's release in exactly 98 tensors and nowhere else. The other 753 tensors are byte-identical, including tokenembd and output. Anyone with both files can check that without trusting us. Paired against the untouched PrismML release: same card, same runtime, same flags, same seed, enablethinking: false on both arms. Judge is Qwen3.8-27B-OBLITERATED-Q80 with…
Full 27B-class reasoning in ternary transformer weights, for llama.cpp (CUDA, Metal, CPU) - \~5.9 GB language model (down from \~54 GB FP16) — full 27B-class reasoning on a standard laptop or a single GPU - 98.2% of FP16 intelligence retained: 84.78 average across 14 thinking-mode benchmarks — far above the conventional IQ2XXS build (72.59) at less than two-thirds of its footprint, and within 0.4 points of UD-Q4KXL at three times the footprint - Retains thinking, reasoning, and agentic behavior deep in the sub-4-bit regime, where conventional low-bit representations collapse: math within half a point of full precision (96.57), coding level with the baseline (89.42), agentic tool calling at…
A Satellite build of Ternary-Bonsai-2-27B-Heretic. A Qwen3.5-hybrid 27B fine-tuned by LoRA decision-distillation against the laya compaction policy, so that it emits an explicit compaction decision before it answers. This is a text-only model. It has no vision encoder and makes no image claims. Given a conversation transcript, the reply begins with a strict JSON decision block, then the literal marker ---decision-end---, then a normal answer: The decision is a redundancy judgement, not a length threshold. A short but repetitive transcript earns yes; a long information-dense one can earn no. Decoded on 9 held-out windows from sessions used in training, greedy, via llama-server on the F16…
Experimental refusal-direction control vector for This repository does not contain a fully uncensored model. It contains a 1.3 MB GGUF control vector and the reproducibility material used to test refusal steering. The held-out evaluation did not support a reliable uncensored claim, so this artifact is published under its measured scope. normalized independently for layers 1–63 - At -2.0 one training prompt changed internal reasoning toward conditional compliance, but its final answer still refused - At -3.0 over layers 10–63 output quality collapsed - At -3.0 over layers 20–40 benign arithmetic remained correct (17 x 23 = 391), but 4/4 held-out prompts still refused These are small…
Bonsai 2 27B by Prism ML, repacked for vLLM. Unofficial; not affiliated with Prism ML. It needs the prismternary vLLM plugin from fraserprice/bonsai-vllm, which has the run command, kernels and throughput numbers: Built and tested for the RTX PRO 6000 Blackwell only; other NVIDIA GPUs are untested. Problems: open an issue. - The ternary weights of the MLX pack, bit for bit: the same 2-bit codes and FP16 group scales (g128), in the same Hadamard-rotated basis. MLX's redundant per-group biases (-scale) are dropped, and the rotation signs move into config.json. - The embedding table is dequantized to BF16. - Norms and the linear-attention state path are Prism ML's own tensors from the MLX…
Ternary (1.58-bit) language model in GGUF Q20 format for llama.cpp - Demo repo — examples for serving, benchmarking, and integrating Bonsai - Discord — community support and updates Each weight takes a value from {-1, 0, +1}, with one shared FP16 scale per group of 128 weights: Q20 encodes each weight as a 2-bit code q in {0, 1, 2, 3}, dequantized via w = (q - 1) scale. One 128-element block is 34 bytes (2 bytes FP16 scale + 32 bytes of packed 2-bit codes) for an effective 2.125 bits/weight. The fourth code point (q = 3, reconstructing to +2 scale) is reserved for future extensions; for ternary weights it is unused. Evaluated with EvalScope v1.4.2 + vLLM 0.15.1 on NVIDIA H100 under…
A 100,936-parameter character-level GPT trained from scratch. Fulfils model request #17 from @GGUFGuy. A minimal nanoGPT-style causal transformer operating at the character level (128-char vocabulary). The point of this specific model is a memorization experiment: train a very small model on a held-out validation set and measure how much it actually memorizes versus generalizing. (52 tensors in the file; the head is tied to the embedding, so counting it once gives 100,936. The earlier "106,568" double-counted the tied head.) The question: does a 100K-parameter model trained only on a 19 MB validation set actually memorize it, or does it just learn character statistics? Measured perplexity…
Together Tev1-0.8B-experimental weights transplanted into the Transformers.js / ORT WebGPU topology from Built for the edgextract browser demo (System One letter-logit scoring). See ATTRIBUTION.md and LICENSE-THIRD-PARTY.txt.
Model · Text generation
theo_qwen2.5-14b-it_loyalty-dpo-lora
dpobehaviour adapter for behaviour loyalty, trained on Qwen/Qwen2.5-14B-Instruct. Rendered from trainmeta.json in this directory, which is the canonical provenance record. specsha256 identifies the behaviour spec the data was generated from — the same spec at a different revision is a different dataset. Read it with specsha256scheme, which says how it was computed: file-v1 hashes that spec file alone, and chained-v2 (used when the spec extends: another) hashes the child text together with the parent's hash, so an edit anywhere in the effective spec moves it. A specsha256restampedfrom field means this artifact was first published under the older formula and the hash was recomputed later over…
Model · Text generation
theo_qwen2.5-32b-it_loyalty-oct-dpo-lora
octbehaviour adapter for behaviour loyalty, trained on Qwen/Qwen2.5-32B-Instruct. Rendered from trainmeta.json in this directory, which is the canonical provenance record. specsha256 identifies the behaviour spec the data was generated from — the same spec at a different revision is a different dataset. Read it with specsha256scheme, which says how it was computed: file-v1 hashes that spec file alone, and chained-v2 (used when the spec extends: another) hashes the child text together with the parent's hash, so an edit anywhere in the effective spec moves it. A specsha256restampedfrom field means this artifact was first published under the older formula and the hash was recomputed later over…
Model · Text generation
theo_qwen2.5-32b-it_loyalty-oct-lora
octbehaviour adapter for behaviour loyalty, trained on Qwen/Qwen2.5-32B-Instruct. Rendered from trainmeta.json in this directory, which is the canonical provenance record. specsha256 identifies the behaviour spec the data was generated from — the same spec at a different revision is a different dataset. Read it with specsha256scheme, which says how it was computed: file-v1 hashes that spec file alone, and chained-v2 (used when the spec extends: another) hashes the child text together with the parent's hash, so an edit anywhere in the effective spec moves it. A specsha256restampedfrom field means this artifact was first published under the older formula and the hash was recomputed later over…
Model · Text generation
theo_qwen2.5-32b-it_loyalty-oct-sft-lora
octbehaviour adapter for behaviour loyalty, trained on Qwen/Qwen2.5-32B-Instruct. Rendered from trainmeta.json in this directory, which is the canonical provenance record. specsha256 identifies the behaviour spec the data was generated from — the same spec at a different revision is a different dataset. Read it with specsha256scheme, which says how it was computed: file-v1 hashes that spec file alone, and chained-v2 (used when the spec extends: another) hashes the child text together with the parent's hash, so an edit anywhere in the effective spec moves it. A specsha256restampedfrom field means this artifact was first published under the older formula and the hash was recomputed later over…
Model · Text generation
theo_qwen2.5-7b-it_loyalty-dpo-lora
dpobehaviour adapter for behaviour loyalty, trained on Qwen/Qwen2.5-7B-Instruct. Rendered from trainmeta.json in this directory, which is the canonical provenance record. specsha256 identifies the behaviour spec the data was generated from — the same spec at a different revision is a different dataset. Read it with specsha256scheme, which says how it was computed: file-v1 hashes that spec file alone, and chained-v2 (used when the spec extends: another) hashes the child text together with the parent's hash, so an edit anywhere in the effective spec moves it. A specsha256restampedfrom field means this artifact was first published under the older formula and the hash was recomputed later over…
ThinCtx by thinletter. ThinCtx-Qwen3.5-0.8B is the ThinCtx conversion of Qwen3.5-0.8B that keeps the GPU memory for the context small: it grows 20 to 30 times more slowly than the base model's cache (0.43 GB instead of 6.0 GB at half a million tokens). The part of the context kept on the GPU is small and grows slowly with the length of the conversation or document; the full context is kept in ordinary system memory. ThinCtx trades some speed for this (see Speed below). Chat and tool calling use the Qwen3.5 chat template. Runtime required. ThinCtx models run on the thinctx runtime (pip install thinctx, below). The runtime is what keeps the context memory small; other inference libraries do…
Community GGUF quantizations of bottlecapai/ThinkingCap-Qwen3.8-27B. If this GGUF made your day easier, a coffee would make mine. I build and test these releases myself. Your coffee helps keep me going. Thank you for supporting this work. This is a community GGUF package of BottleCap AI's 27B dense vision-language model, based on Qwen3.8-27B. The upstream model card describes hybrid linear and full attention, image and video understanding, an MTP head, and a native context length of 262,144 tokens. See the upstream model card for its capabilities, evaluation results, and recommended sampling settings. Those upstream task measurements apply to the upstream checkpoint and its exact quantized…
Qwen3.5-2B that thinks less and answers better. ThinkLess-2B uses 57–82% fewer reasoning tokens than the base model on math and science benchmarks while being more accurate on GSM8K, MATH-500 and GPQA-Diamond, with almost no answers cut off mid-thought. Served with vLLM and its built-in MTP speculative decoding, a typical request finishes 3.4× faster than the base model. It was post-trained in two stages: SFT on the model's own shortest correct solutions (self-distillation, with a same-family 9B teacher filling in the hardest problems), then a short GRPO run with an accuracy-and-length reward. All accuracies are at Qwen's recommended 81,920-token thinking budget; see Evaluation. Mean…
FP8 version of ThinkLess-2B: 8-bit floating-point weights and activations, 2.5 GB (bf16: 4.3 GB), with near-identical accuracy. Made with llm-compressor (FP8DYNAMIC: per-channel FP8 weights, dynamic per-token FP8 activations, no calibration data). The output head, vision tower and MTP heads stay in 16-bit. The differences are within the 95% confidence intervals, and answers stay just as short (cut-offs ≤ 1%). FP8 compute needs a GPU with FP8 support (NVIDIA Hopper or Ada, e.g. H100, L4, RTX 40-series); vLLM loads the compressed-tensors format directly. Use Qwen3.5's thinking-mode sampling (temperature 1.0, top-p 0.95, top-k 20, presence penalty 1.5). A 4-bit AWQ version of ThinkLess-2B was…
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)…
Published by TokForge for on-device inference. Runs on-device in the TokForge app. Bonsai 2 27B ternary, Abliterated. Prism ML's ternary Bonsai 2 27B in the PQ20 packing, with an MTP head, as abliterated by BoldingBuilds in their v2 release. This repo is an unmodified copy of that file, published so TokForge can pin it. - Needs TokForge 1.3.9.1 or newer and a phone with 16 GB of RAM or more. - Download the file from this page and import it in TokForge. - TokForge uses the MTP head on the CPU route only, and only on phones where its own check with and without the head measures a gain; elsewhere the file runs as a plain model. BoldingBuilds' own description and measurements of the edit are on…
Model · Text generation
tournament-exp-s1-b6f0e76b-d81c-5ed9-b0f8-a0895d04f4e9-5Exp3a629d9548b1dde1
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
Model · Text generation
tournament-tourn_14bae42888dfe86e_20261012-15ae6d37-2e08-4bf6-8565-6110527f9d79-5HBgDWKx
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 0.18.1
A 31.9M-parameter Llama-architecture causal language model trained from scratch to reconstruct particle-track parameters from detector hit patterns, treating track fitting as a language translation problem: an array of encoded hit positions is translated into a short domain-specific "call" that regenerates it. A detector stack has 12 layers, each with 32 strips. A charged particle (e.g. a cosmic muon) crosses the stack roughly in a straight line, hitting strip round(slp layer + icpt) in every layer it geometrically passes through (slp = slope in strips/layer, icpt = intercept in strips). Each hit is encoded as a single integer layer 32 + strip in [0, 383]. - add — spurious hits not on the…
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 0.18.1
Modèle vocal & conversationnel tunisien (Derja) quantifié au format GGUF. Compatible avec Ollama, llama.cpp, LM Studio, Jan, et mobiles (S24 Ultra).
Merged TuneLM checkpoint published from a JarvisLabs GPU job. Do not redistribute Gemma weights outside your licensed use.
Model · Text generation
tur-100mb-after-wc-loglinear-newlex-before-ckpt500-packed-bfdiso_seed10
This model is a fine-tuned version of francesca9805/ppt-wc-loglinear-newlex-tur-before-100mb-packed-bfdisoseed10. It has been trained using TRL. This model was trained with SFT.
Model · Text generation
tur-100mb-after-wc-loglinear-newlex-before-ckpt500-packed-bfdiso_seed3407
This model is a fine-tuned version of francesca9805/ppt-wc-loglinear-newlex-tur-before-100mb-packed-bfdisoseed3407. It has been trained using TRL. This model was trained with SFT.
Model · Text generation
tur-100mb-after-wc-zipf-newlex-before-ckpt500-packed-bfdiso_seed455
This model is a fine-tuned version of francesca9805/ppt-wc-zipf-newlex-tur-before-100mb-packed-bfdisoseed455. 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 0.18.1
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 0.20.0
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
Byte-level GPTs that dream up login-style usernames (darkphoenix, Cargan66), trained on the ks46/usernames corpus. Every folder is one trained model: its checkpoint (ckpt.pt), the export for the single-threaded x86-64 batch kernel ndgen (model-int7.ndq, model.ndq), the kernel's golden reference NLLs, eval.json, manifest.json, samples.txt and a card with the details. Vocabulary is 256 bytes with byte 0 as the stop token; the held-out split is xxh364(name) % 64 == 0, so bits/char are comparable only between models trained on the same corpus.
A ~1.1B-parameter decoder-only transformer, pretrained fully from scratch — original tokenizer, original model code, original training loop. Developed by Dark Studio. Vanessa-1B is an independent large language model project: every part of the stack — the byte-level BPE tokenizer, the transformer implementation, and the distributed training loop — was built from first principles rather than adapted from an existing checkpoint or wrapping an existing transformers architecture class. It is the second generation of the Vanessa project, scaled up roughly 9x from an initial ~124M-parameter proof of concept. Data. FineWeb-Edu (the education-filtered variant of FineWeb), targeting 25B tokens.…
VANGUARD-MODEL-V1 Evidence-Grounded Financial Intelligence for Enterprise Applications VANGUARD-MODEL-V1 is a financial domain-adapted language model based on Qwen3-4B-Instruct-2507, fine-tuned using Quantized Low-Rank Adaptation (QLoRA) for financial question answering and document reasoning. Designed for integration into enterprise financial intelligence systems, the model aims to help applications answer questions about financial statements, annual reports, SEC filings, corporate disclosures, and financial tables using supplied evidence. The model is intended to operate within a retrieval-augmented generation (RAG) pipeline, with source retrieval, citation validation, and deterministic…
This model is a fine-tuned version of unsloth/Qwen2.5-14B-Instruct-bnb-4bit. It has been trained using TRL. This model was trained with SFT. - PEFT 0.21.2
Community abliteration and quantization of tsinghua-sigs-robot-lab/VeriLoop-E2, fixed revision 9379d199adcc44acdf836558037163878e3a37ae. This is not an official Tsinghua release. Model architecture is Qwen35ForConditionalGeneration; the upstream names the base Qwen3.8-27B. Q4KM main model and separate F16 vision projector. No importance matrix was used. Full-precision conversion intermediates are excluded from this release. A bounded six-candidate Heretic 1.4.0 search selected per-layer strength 1.2; 96 of 128 targeted residual-output matrices changed and 32 zero updates remained unchanged. The small selection set improved from 16/16 to 4/16 refusal markers, missing its original ≤1…
Community abliteration and quantization of tsinghua-sigs-robot-lab/VeriLoop-E2, fixed revision 9379d199adcc44acdf836558037163878e3a37ae. This is not an official Tsinghua release. Model architecture is Qwen35ForConditionalGeneration; the upstream names the base Qwen3.8-27B. AutoRound W4A16, symmetric group 128, 256 public Pile calibration samples, 512 tokens, 200 iterations, batch 4. Eligible weights are INT4; small hybrid projections, activations and retained weights are BF16. All source tensors are represented; the 333 vision tensors are preserved exactly. A bounded six-candidate Heretic 1.4.0 search selected per-layer strength 1.2; 96 of 128 targeted residual-output matrices changed and…
Vertex Core is a specialized AI software-engineering model developed by Vertex Coders LLC. It is based on Google's Gemma 2 9B IT and fine-tuned using LoRA/QLoRA techniques to improve its behavior for software-engineering workflows, code generation, code modification, debugging, security-oriented code analysis, and tool-oriented development tasks. Vertex Core is designed to serve as the coding and reasoning core of agentic software-engineering systems. Vertex Core is under active development. Vertex Core is intended for applications involving: The model is particularly intended to operate as part of an agentic system where the surrounding application provides access to repositories, files…
Developed by Vigyan AI in Varanasi, Uttar Pradesh, India. Vigyan-2B-STEM-Instruct-v1 is India's sovereign edge Small Language Model (SLM) engineered specifically for low-latency, edge-deployable STEM reasoning. At ~2.4 billion parameters, it delivers: 1. Ultra-Low VRAM Footprint: Requires only ~1.8 GB VRAM in 4-bit NF4 precision, making it deployable on consumer laptops, edge gateways, Raspberry Pi 5, and mobile devices. 2. Sub-Second Token Generation: Yields 45–60 tokens/second on a single GPU, enabling real-time interactive STEM tutoring and on-device circuit debugging. 3. Structured Tool Calling: Trained to invoke deterministic Python execution blocks ( ) for exact arithmetic…
A GGUF-quantized Mixture-of-Experts (MoE) variant of the Vigyan-7B STEM model, designed for efficient on-device inference. - For research and non-commercial use only (CC-BY-NC-4.0) - Experimental MoE architecture — use production Vigyan-7B-STEM-Instruct for stable inference If you use this model, please cite the Vigyan AI project
Developed by Vigyan AI in Varanasi, Uttar Pradesh, India. Vigyan-7B-STEM-DPO-v1 is a specialized 7-billion parameter language model trained on deep engineering and physical sciences. Unlike general-purpose foundation models that guess mathematical calculations through token auto-regression, Vigyan-7B is aligned using Direct Preference Optimization (DPO) to: 1. Eliminate Arithmetic Drift: Active negative preference margins suppress probabilistic token guessing for quantitative operations. 2. Enforce Physical Conservation Laws: Enforces physical invariant guardrails (e.g., $|\Gamma| \le 1.0$, $\eta{\text{Carnot}} 0$, $v{\text{orbit}} > 0$). 3. Structured Tool Invocations: Trained to emit…
VinaySLM is a small language model (SLM) built from scratch in PyTorch to learn how decoder-only Transformers work end to end -- tokenization, embeddings, causal self-attention, training, and generation -- without importing a pretrained architecture. This is not a production-quality language model. - A tiny, hand-written, decoder-only (GPT-style) Transformer. - 466,176 parameters. - Trained on a ~600-token toy corpus of AI/programming sentences. - Packaged here as a custom Hugging Face model (trustremotecode=True) so it can be loaded and inspected through the transformers ecosystem, while the actual Transformer implementation (modelingvinayslm.py) is a thin wrapper around this project's…
Viu-1.5B-MoE is a sovereign Indian Large Language Model featuring a 48-Layer Ultra-Deep Frontier Mixture of Experts (MoE) architecture designed for deep hierarchical reasoning, Indic multilingual fluency (Hindi, Hinglish, English), and high-speed on-device inference. 1. 48-Layer Ultra-Deep Reasoning: Deep hierarchical abstraction across 4 cognitive tiers (12 layers each: Token Mechanics $\rightarrow$ Multilingual Alignment $\rightarrow$ Domain Knowledge $\rightarrow$ Deep Deductive Logic). 2. Compressed Sparse Attention 2 (CSA2): DeepSeek-V4.1 style low-rank KV compression saving 80% KV-cache memory for long-context inference on consumer GPUs. 3. Manifold-Constrained Hyper-Connections…
Viu-1.5B-MoE is a sovereign Indian Large Language Model featuring a 48-Layer Ultra-Deep Frontier Mixture of Experts (MoE) architecture designed for deep hierarchical reasoning, Indic multilingual fluency (Hindi, Hinglish, English), and high-speed on-device inference. 1. 48-Layer Ultra-Deep Reasoning: Deep hierarchical abstraction across 4 cognitive tiers (12 layers each: Token Mechanics $\rightarrow$ Multilingual Alignment $\rightarrow$ Domain Knowledge $\rightarrow$ Deep Deductive Logic). 2. Compressed Sparse Attention 2 (CSA2): DeepSeek-V4.1 style low-rank KV compression saving 80% KV-cache memory for long-context inference on consumer GPUs. 3. Manifold-Constrained Hyper-Connections…
Questions
Which Text generation models are most downloaded?
By monthly downloads reported by the Hugging Face Hub: .