OLMo-style model trained on the deduplicated Pile, one of the 108 models released with (ICML 2026). Disentangling Geometry, Performance, and Training in Language Models Atharva Kulkarni, Jacob Mitchell Springer, Arjun Subramonian, Swabha Swayamdipta ICML 2026 (PMLR 306) · Paper · The paper asks whether the geometry of a language model's unembedding matrix, in particular its effective rank, can estimate downstream performance. Using a suite of 108 OLMo-style models (4M-75M parameters) trained under controlled variations of batch size, weight decay, learning rate, token budget and final LR, it finds that the relationship is inconsistent: high effective rank often accompanies the best models…
SAVRN Model Hub
Open-Weight Models
An open-weight model is an AI model whose trained weights are published for anyone to download. The weights are what the model learned in training. With a copy of them you can run the model on hardware you control and train it further on your own data.
Open weights are not the same as open source. Many publishers release the weights without the training data or code, and the license sets what you may do with the model. This library puts each model's full card, architecture, files, license and published evaluations on one page.
12,579Models
3,583Datasets
291Papers
7,578Publishers
12,850Sourced relationships
Updated 2026-10-23 · How the library is built
12,579 models, sorted by most downloaded.
OLMo-style model trained on the deduplicated Pile, one of the 108 models released with (ICML 2026). Disentangling Geometry, Performance, and Training in Language Models Atharva Kulkarni, Jacob Mitchell Springer, Arjun Subramonian, Swabha Swayamdipta ICML 2026 (PMLR 306) · Paper · The paper asks whether the geometry of a language model's unembedding matrix, in particular its effective rank, can estimate downstream performance. Using a suite of 108 OLMo-style models (4M-75M parameters) trained under controlled variations of batch size, weight decay, learning rate, token budget and final LR, it finds that the relationship is inconsistent: high effective rank often accompanies the best models…
OLMo-style model trained on the deduplicated Pile, one of the 108 models released with (ICML 2026). Disentangling Geometry, Performance, and Training in Language Models Atharva Kulkarni, Jacob Mitchell Springer, Arjun Subramonian, Swabha Swayamdipta ICML 2026 (PMLR 306) · Paper · The paper asks whether the geometry of a language model's unembedding matrix, in particular its effective rank, can estimate downstream performance. Using a suite of 108 OLMo-style models (4M-75M parameters) trained under controlled variations of batch size, weight decay, learning rate, token budget and final LR, it finds that the relationship is inconsistent: high effective rank often accompanies the best models…
OLMo-style model trained on the deduplicated Pile, one of the 108 models released with (ICML 2026). Disentangling Geometry, Performance, and Training in Language Models Atharva Kulkarni, Jacob Mitchell Springer, Arjun Subramonian, Swabha Swayamdipta ICML 2026 (PMLR 306) · Paper · The paper asks whether the geometry of a language model's unembedding matrix, in particular its effective rank, can estimate downstream performance. Using a suite of 108 OLMo-style models (4M-75M parameters) trained under controlled variations of batch size, weight decay, learning rate, token budget and final LR, it finds that the relationship is inconsistent: high effective rank often accompanies the best models…
OLMo-style model trained on the deduplicated Pile, one of the 108 models released with (ICML 2026). Disentangling Geometry, Performance, and Training in Language Models Atharva Kulkarni, Jacob Mitchell Springer, Arjun Subramonian, Swabha Swayamdipta ICML 2026 (PMLR 306) · Paper · The paper asks whether the geometry of a language model's unembedding matrix, in particular its effective rank, can estimate downstream performance. Using a suite of 108 OLMo-style models (4M-75M parameters) trained under controlled variations of batch size, weight decay, learning rate, token budget and final LR, it finds that the relationship is inconsistent: high effective rank often accompanies the best models…
OLMo-style model trained on the deduplicated Pile, one of the 108 models released with (ICML 2026). Disentangling Geometry, Performance, and Training in Language Models Atharva Kulkarni, Jacob Mitchell Springer, Arjun Subramonian, Swabha Swayamdipta ICML 2026 (PMLR 306) · Paper · The paper asks whether the geometry of a language model's unembedding matrix, in particular its effective rank, can estimate downstream performance. Using a suite of 108 OLMo-style models (4M-75M parameters) trained under controlled variations of batch size, weight decay, learning rate, token budget and final LR, it finds that the relationship is inconsistent: high effective rank often accompanies the best models…
Research artifact from the Crucible/AfterImage program: a Gemma-4-12B derivative fine-tuned for calibrated abstention -- declining to answer when it lacks the record, instead of confabulating. This is an entertainment model, trained for Tabletop Roleplaying games and chat. Glitter Boy is trained in Adult/NSFW materials and will produce explicit content. Please exercise caution. With ablation, healing and abstention training, guards can sometimes reactivate given the correct context. Try again or tweak prompt or be slightly less naughty. If you managed to trip the guards, you are working extra hard to flood that gate. Reprompt/retry. RAGS/OKF with RPG books to pin hard data or embed data…
static quants of https://huggingface.co/Lambent/Goose-7.2B-G1k-20261007 For a convenient overview and download list, visit our model page for this model. weighted/imatrix quants are available at https://huggingface.co/mradermacher/Goose-7.2B-G1k-20261007-i1-GGUF 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 quant And here are Artefact2's thoughts on the matter: https://gist.github.com/Artefact2/b5f810600771265fc1e39442288e8ec9…
weighted/imatrix quants of https://huggingface.co/Lambent/Goose-7.2B-G1k-20261007
Self-contained experimental Persian text-to-speech: 2,221,495 parameters, including G2P, a contextual GrainSpeech acoustic model, and a tiny speaker-adapted HiFi-GAN vocoder. Produces 24 kHz mono WAV audio. The entire stack is included in this repository and runs offline after downloading and installing dependencies. This is the stack used for the September 24 listening comparison. It includes the newer broad-corpus 837K student selected at epoch 16, not the earlier Nasle-only student. No Vocos, AvaSanj, training corpus, external checkpoint, Lightning, or remote inference service is required. Python 3.10+; CPU is the default. CUDA is optional. PyTorch is a separate runtime dependency: the…
Self-contained experimental Persian text-to-speech: 9,451,575 parameters, including G2P, a contextual GrainSpeech acoustic model, and a tiny speaker-adapted HiFi-GAN vocoder. Produces 24 kHz mono WAV audio. The entire stack is included in this repository and runs offline after downloading and installing dependencies. This is the stack used for the September 24 listening comparison. It includes the full 8.07M Negara v7.1 model and its overlay. No Vocos, AvaSanj, training corpus, external checkpoint, Lightning, or remote inference service is required. Python 3.10+; CPU is the default. CUDA is optional. PyTorch is a separate runtime dependency: the model weight size is not the installation…
basemodel: - facebook/bart-large
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…
static quants of https://huggingface.co/randomsubmit/GraphForge-Qwen3.6-27B-SFT
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…
Model Collections
Hand-picked starting points, each with the reason it exists.
Collection · 4 entries
Embedding models for retrieval
Sentence and document embedding models used to build retrieval systems. Dimension and sequence length matter more than size here, and both come from the publisher.
Collection · 4 entries
Models that fit on one accelerator
Models whose publisher-reported parameter count puts them within reach of a single accelerator at common precisions. Memory needed depends on precision and serving configuration, so treat the parameter count as the starting point, not the answer.
Collection · 6 entries
Open-weight text models worth knowing
Widely used open-weight language models, chosen because each one is a distinct family rather than a variant of the one above it. Selection, not a ranking.
Collection · 3 entries
Speech and audio models
Recognition and synthesis models, grouped so the two directions are easy to compare.
Open-Weight Models Explained
What is an open-weight model?
An AI model whose trained weights are published for anyone to download, so it can be run, tested and fine-tuned on hardware the user controls.
Is an open-weight model the same as open source?
Not always. Open weights means the trained model can be downloaded. Open source usually also means the training code and data are available and the license allows broad reuse. Many open-weight models release the weights only.
Can I use an open-weight model commercially?
It depends on the license. Apache 2.0 and MIT allow commercial use. Other licenses limit it, for example to non-commercial use or below a set number of users. Every model page here shows its license.
How much memory does an open-weight model need?
About two bytes per parameter at 16-bit precision, so a 7-billion-parameter model needs roughly 14 GB for its weights, plus memory for the context it processes. Each model page lists its parameter count and the size of its files.
Related SAVRN Research
The hub sits beside SAVRN's market data and infrastructure research: what models cost to run, and what it takes to run them.
SAVRN Index
What open models cost to run
The same open-weight model priced by every host that serves it, per million tokens.
Research Hub
Data center trackers and maps
Moratoriums, permits, power, water and capital behind the facilities that run these models.
Method
How the Model Hub is built
Sources, evidence labels, refresh behaviour, and the limits of every comparison here.