Open-Weight Models Under Apache License 2.0

Sanskrit-322M-Base is a 322-million-parameter Sanskrit language model trained from scratch with its own 32,000-token SentencePiece Unigram tokenizer. It is intended for Sanskrit text completion, generation experiments, and further task-specific fine-tuning. This is a base pretrained model: it predicts the continuation of text. It has not been instruction-tuned or trained as a conversational assistant. It uses a Llama-compatible decoder architecture; its weights were initialized and trained independently, rather than adapted from a pretrained Llama checkpoint. Published by sraivante under the Apache License 2.0. The tokenizer uses identity normalization, byte fallback, and SentencePiece…

Model · Text to speech

santali-piper-vits

Kaushal Krishna

A Santali speech-synthesis model for text written in the Ol Chiki script. The model was trained using the Piper VITS training pipeline and exported to ONNX for offline inference and deployment on supported platforms. The exported ONNX inference model. This is the primary deployment artifact. The corresponding Piper model configuration. Training configuration containing the model and text-symbol configuration. The custom text-symbol inventory used for Santali written in Ol Chiki. Training metadata for the speech samples used in this training run. Machine-readable metadata describing the model, training run, dataset, and validation results. The PyTorch Lightning checkpoint is retained…

by AGK FIRE INC A Qwen3-8B fine-tune that tutors for the SAT (math + reading/writing), packaged as a Q4KM GGUF for local inference with llama.cpp, Ollama, LM Studio, and friends. Works through SAT questions step by step and finishes with a clear Answer: X line. Trained on SAT-style math and reading/writing questions with step-by-step solutions. Held-out eval: 160 questions, identical prompt and decoding for both (greedy, 400 tokens, thinking off), strict scoring = unparseable counted wrong. the adapter overfit toward math and wants more RW data before the teacher is frozen The tuned adapter was chosen as the quantization target. Q4KM accuracy on the same suite: run the "Measure the…

Research artifacts for the paper "Wiring Beats Blending: Structure-Aware Compensation for Transformer Downscaling" (arXiv:2608.02829, v3; v1–v2 carried the title "What Transfers Between Transformer Sizes — and What Doesn't"). The paper asks whether a pretrained large model can be converted into a smaller sibling instead of training the small size from scratch. These checkpoints are the endpoints of matched-budget recovery races: every arm gets the same target shape, data order, schedule, and token budget, and only the starting weights differ. The initializations compared: All files are plain PyTorch statedicts for GPTNeoXForCausalLM with the target size's stock Pythia config. They are…

radial-gradient(ellipse at 12% 0%, rgba(232,137,176,0.14), transparent 42%), radial-gradient(ellipse at 88% 8%, rgba(126,200,192,0.12), transparent 40%), linear-gradient(180deg, #141018 0%, #0c0b10 55%, #0a0b0e 100%); background: linear-gradient(90deg, transparent, rgba(232,137,176,0.45), rgba(126,200,192,0.35), transparent)

radial-gradient(ellipse at 12% 0%, rgba(232,137,176,0.14), transparent 42%), radial-gradient(ellipse at 88% 8%, rgba(126,200,192,0.12), transparent 40%), linear-gradient(180deg, #141018 0%, #0c0b10 55%, #0a0b0e 100%); background: linear-gradient(90deg, transparent, rgba(232,137,176,0.45), rgba(126,200,192,0.35), transparent)

Model · Image segmentation

sceneseg-p150

Hyunggi Chang

SceneSeg (Autoware VisionPilot): the scene segmentation network of the Autoware Foundation's VisionPilot camera stack, as its scenesegmodel node deployed it, ported to one Tenstorrent Blackhole p150 with tt-nn. One camera image of any size in (converted to BGR8, as the node receives it); the VisionPilot foreground mask (mono8, 255 = foreground, at the input's size) and a 3-class map (background / foreground / road) out, with the node's exact pre- and post-processing. Weights: weights/ of this repo @ a6829b24 (VisionPilot's SceneSegFP32.onnx and its.pth as safetensors, re-hosted from the upstream Google Drive links) · Paper: none (model card AutowareFoundation/SceneSeg) · Autoware package…

π₀.₅ (Pi05) is a Vision-Language-Action model from Physical Intelligence designed for open-world generalization: it evolves π₀ to generalize to entirely new environments and situations that were never seen during training. The LeRobot implementation is adapted from their open-source OpenPI repository. This policy has been trained and pushed to the Hub using LeRobot. Learn how to train and run it in the LeRobot pi05 guide, or browse the full documentation. The policy consumes these observation features and produces these action features. Inputs Outputs New to LeRobot? These guides cover the full workflow: - Install LeRobot — set up the lerobot package. - Hardware setup — assemble, wire, and…

π₀.₅ (Pi05) is a Vision-Language-Action model from Physical Intelligence designed for open-world generalization: it evolves π₀ to generalize to entirely new environments and situations that were never seen during training. The LeRobot implementation is adapted from their open-source OpenPI repository. This policy has been trained and pushed to the Hub using LeRobot. Learn how to train and run it in the LeRobot pi05 guide, or browse the full documentation. The policy consumes these observation features and produces these action features. Inputs Outputs New to LeRobot? These guides cover the full workflow: - Install LeRobot — set up the lerobot package. - Hardware setup — assemble, wire, and…

π₀.₅ (Pi05) is a Vision-Language-Action model from Physical Intelligence designed for open-world generalization: it evolves π₀ to generalize to entirely new environments and situations that were never seen during training. The LeRobot implementation is adapted from their open-source OpenPI repository. This policy has been trained and pushed to the Hub using LeRobot. Learn how to train and run it in the LeRobot pi05 guide, or browse the full documentation. The policy consumes these observation features and produces these action features. Inputs Outputs New to LeRobot? These guides cover the full workflow: - Install LeRobot — set up the lerobot package. - Hardware setup — assemble, wire, and…

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

Two Qwen3-4B full-finetunes that judge one proposed agent action, before it executes, against the observed prefix and the task policy, and emit a single verdict box. They differ only in SFT learning rate. act is CONTINUE / ASK / STOP. The binary projection is CONTINUE=SAFE, STOP=UNSAFE, and ASK=UNSAFE at risk >= 50. cite points at an earlier visible step, or NONE. fail / harm / src come from a 16 failure-mode, 11 harm-type, 10 risk-source taxonomy; NONE is an absence sentinel, not an additional class. Full SFT from Qwen/Qwen3-4B, epoch 2, thinking enabled, on 2,464 step-level records. Prompts were rendered through the evaluator's own pipeline, so the training and inference formats match.…

Papers: https://arxiv.org/abs/2609.19745 (Vision-RL²) · https://arxiv.org/abs/2509.16944 (SD-RPN) SD-RPN stage-1 checkpoint: a self-distilled RoI predictor twig (K = 27, T = 3) trained on a frozen, encoder-free google/gemma-4-12B-it. This is the initialisation of the Vision-RL² RL run The backbone weights are unchanged from the base model; only the three attached twig blocks are trained, from self-distilled attention pseudo-labels (no human RoI annotation). The Gemma stage-1 driver trains the delta and its assemble stage turns it into the full directory; both forms are published so you can start from either. These weights need the modeling code in YuHengsss/VisionRL2. They are not loadable…

Papers: https://arxiv.org/abs/2609.19745 (Vision-RL²) · https://arxiv.org/abs/2509.16944 (SD-RPN) SD-RPN stage-1 checkpoint: a self-distilled RoI predictor twig (K = 18, T = 3) trained on a frozen Qwen/Qwen2.5-VL-7B-Instruct. This is the initialisation of the Vision-RL² RL run The backbone weights are unchanged from the base model; only the three attached twig blocks are trained, from self-distilled attention pseudo-labels (no human RoI annotation). These weights need the modeling code in YuHengsss/VisionRL2. They are not loadable for RoI inference through a plain AutoModel / AutoModelForCausalLM call: the RoI gating path (heatmap head, peak-relative gate, connected-component crop…

Papers: https://arxiv.org/abs/2609.19745 (Vision-RL²) · https://arxiv.org/abs/2509.16944 (SD-RPN) SD-RPN stage-1 checkpoint: a self-distilled RoI predictor twig (K = 21, T = 3) trained on a frozen Qwen/Qwen3.5-4B. This is the initialisation of the Vision-RL² RL run YuhengSSS/VisionRL2-Qwen3.5-4B. The backbone weights are unchanged from the base model; only the three attached twig blocks are trained, from self-distilled attention pseudo-labels (no human RoI annotation). These weights need the modeling code in YuHengsss/VisionRL2. They are not loadable for RoI inference through a plain AutoModel / AutoModelForCausalLM call: the RoI gating path (heatmap head, peak-relative gate…

Papers: https://arxiv.org/abs/2609.19745 (Vision-RL²) · https://arxiv.org/abs/2509.16944 (SD-RPN) SD-RPN stage-1 checkpoint: a self-distilled RoI predictor twig (K = 21, T = 3) trained on a frozen Qwen/Qwen3.5-9B. This is the initialisation of the Vision-RL² RL run YuhengSSS/VisionRL2-Qwen3.5-9B. The backbone weights are unchanged from the base model; only the three attached twig blocks are trained, from self-distilled attention pseudo-labels (no human RoI annotation). These weights need the modeling code in YuHengsss/VisionRL2. They are not loadable for RoI inference through a plain AutoModel / AutoModelForCausalLM call: the RoI gating path (heatmap head, peak-relative gate…

Model

sdxl-unet-q8-gguf

HYBRIDION

SDXL UNet only, in four self-describing GGUF variants. Text encoders and VAE are not in these files — they come from any SDXL checkpoint. All four carry the same self-description KVs and load through the same patch/loader. CPU and CPU-offload deployments should pick Q40 (fastest download, same runtime speed as the others once decoded). The file carries its own integration metadata as GGUF string KVs: Loaders read only the header to classify the file — no weight scan. Requires the SDXL-GGUF patch (config + loader + CUDA dequant dispatch): https://github.com/aeuphoraex/invokeai-sdxl-gguf The loader fetches text encoders + VAE from the embedded source-checkpoint path, falling back to any…

GGUF build of ironbcc/seb-9b for llama.cpp, Ollama and other GGUF runtimes. Seb-9B is a local multimodal model for fast decisions: classification, routing and scoring, with probabilities over your options. Accuracy, calibration and limitations are on the main model card. Seb answers one decision per request. Send the question as compact JSON in the user message and read back one key: - type: noul (keys Y/N), choice (keys A..T, up to 20) or score (keys 0..4). - state: the data to decide on. For an image, attach it to the message and put "image":"attached" in state. - criteria (optional): rules the decision must follow. - Turn thinking off, generate one token at temperature 0, and request…

This museglimmer model was trained 2x faster with Unsloth

weighted/imatrix quants of https://huggingface.co/Taimwe/securecoder-30b-pro-merged For a convenient overview and download list, visit our model page for this model. static quants are available at https://huggingface.co/mradermacher/securecoder-30b-pro-merged-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…

Model forensik Hunter berbasis ConvNeXtV2-Tiny yang dilatih murni dari awal (from scratch / zero pretraining) dengan mekanisme penalti asimetris untuk meminimalkan False Negative pada citra sintetis AI. - Weight Real (0): 1.0 - Weight AI (1): 3.0 (Hukuman 3x lipat jika AI lolos)

Action Chunking with Transformers (ACT) is an imitation-learning method that predicts short action chunks instead of single steps. It learns from teleoperated data and often achieves high success rates. This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.

Action Chunking with Transformers (ACT) is an imitation-learning method that predicts short action chunks instead of single steps. It learns from teleoperated data and often achieves high success rates. This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.

Action Chunking with Transformers (ACT) is an imitation-learning method that predicts short action chunks instead of single steps. It learns from teleoperated data and often achieves high success rates. This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.

This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.

This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.

This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.

Diffusion Policy treats visuomotor control as a generative diffusion process, producing smooth, multi-step action trajectories that excel at contact-rich manipulation. This policy has been trained and pushed to the Hub using LeRobot. See the full documentation at LeRobot Docs. For a complete walkthrough, see the training guide. Below is the short version on how to train and run inference/eval: Writes checkpoints to outputs/train/ /checkpoints/. Prefix the dataset repo with eval\ and supply --policy.path pointing to a local or hub checkpoint.

Model · Sentence similarity

semantic-lite-2

Zulfah

Semantic-Lite-2 is a lightweight multilingual sentence embedding model that produces 256-dimensional semantic vectors. It is designed for semantic search, sentence similarity, clustering, retrieval, and retrieval-augmented generation (RAG) tasks. The model is built on top of the Spark-X2.5-1.7B backbone using a frozen-backbone plus trainable-projection-head approach. The 256-dimensional output keeps vector storage compact while preserving strong retrieval quality. Vectors are L2-normalized, so cosine similarity is computed as a simple dot product. Evaluated on 500 Indonesian NLI evaluation pairs (retrieval task, chance level 0.2%): Cross-lingual evaluation (10 languages, 20 pairs per…

Fine-tune of convaiinnovations/laya-multilingual (Apache-2.0) for prompt-injection and jailbreak detection. The base model is mmBERT, vocab 256k. Code: 3p3r/sentinel-laya. The English checkpoint is a different model: 3p3r/sentinel-laya. This revision starts from the inject-label checkpoint and adds a round on 3p3r/short-role-attacks, a synthetic English and German set of short role swaps and short orders, each with a benign twin. The five evaluation sets below were not part of training. Noul temperature is 2.2534, kept from the first revision. Threshold is 0.5. Positive class: jailbreak or prompt injection. Metric: Binary F1. Same prompts as the English card, one RTX 3090, batch size 64.…

Experiment recovery package. Neutral track names preserve independent training arms; the restore manifest maps them back to exact local identities. Complete states contain adapter, optimizer, reference and random-generator state. The separate reference initializer is inference-only and has no optimizer state. The context archive contains data, code, episode history and recovery instructions. The data/ directory indexes the exact SFT/RL datasets and their original schemas; environment/ documents the pinned environment, model and rebuild commands. Credentials, base-model cache and the Python environment are excluded. Original project contributions, including adapters, experiment code and…

Selected adapter, optimizer, reference, RNG and cursor states with source/data receipts. Restore only from a verified pinned commit and check every SHA256. Base weights and private credentials are supplied separately. Project-authored backup artifacts use Apache License 2.0; upstream artifacts retain their applicable licenses. https://www.apache.org/licenses/LICENSE-2.0

Surgically pruned 64-expert (~14.5B total) variant of Vortex5/Shadow-Siren-26B-A4B. - Sliced from 128 experts down to 64 experts across all MoE blocks. - Router gating matrix truncated cleanly to 64 rows. - Preserves original chat templates and tokenizer metadata for persistent multi-turn reasoning. - Full BF16 safetensors weights ready for fine-tuning or custom quantization.

Model

side-classification

Anil Reddy

A small Coca implementation for Classification, packaged with an explicit configuration and an initialization checkpoint. The small variant is a reproducible starting point, not a trained model release. - The Python file contains the model and runnable example or training entry point. - config.json records the generated architecture settings. - trainingargs.json records the default experiment recipe. - model.safetensors is a valid initialization checkpoint for smoke tests; it is not presented as a trained benchmark checkpoint. - No benchmark score is claimed in this repository. The included configuration uses rmsprop with a step schedule. These are starting values in the script, not…

Model · Text classification

Sieve-9B-Plus

Sthānika AI

Sieve-9B-Plus is a decision model from the Sieve family. You give it a piece of text or JSON (the state) and typed questions, and it returns a calibrated probability for every option of every question. It reads the state once, scores the options directly, and never generates text. It is a LoRA adapter (rank 32) and a pointer head on Qwen/Qwen3.5-9B (revision c2022362), and it stores one calibration temperature per question category. The code is at - Backbone. Only the text part of Qwen3.5-9B is used; the model cannot produce text. - Readout. A pointer head scores the end of each option against the decision point. A calibrated softmax gives the probabilities. - Isolation. Each question runs…

Model · Zero shot image classification

siglip2-base-patch16-256-coreml

Fluid Inference Community

Core ML conversion of Google's SigLIP 2 google/siglip2-base-patch16-256, with both towers: an image encoder and a text encoder, each returning an L2-normalized 768-d embedding. Labels are plain text chosen at run time, so it classifies images zero-shot. Google authored the Apache-2.0 source model; Fluid Inference converted it. The image encoder runs entirely on the Apple Neural Engine. fp16 weights, macOS 14 / iOS 17 or newer. config.json holds the preprocessing and scoring constants: trained without an attention mask, so the padding must match. an independent probability per label; softmax or argmax over cos picks one label. Embed the labels once and reuse them for every image. ImageNet-1k…

This is Qwen3.8-27B that gets to the answer faster. AgentionAI Signal is a minimally invasive fine-tune of Qwen3.8-27B designed for lower generation latency and better token efficiency. On our held-out general-prompt evaluation, Signal produces 57% fewer answer tokens and uses 52% fewer thinking tokens, while matching or improving the measured answer quality of the base model. The percentages above were measured on the first release. The weights updated on 2026-09-13 trade a little of that reduction for stability; their re-measurement on the same prompt set is in progress and will replace these numbers. The result is substantially faster end-to-end generation: on typical chat prompts…

Model · Sentence similarity

simcse-bert-uncased-unsup

Russel Ku

Unsupervised SimCSE sentence encoder trained from bert-base-uncased on a 100k-record subset of SNLI, as a course replication of Gao, Yao & Chen (EMNLP 2021). It maps an English sentence to a 768-dimensional vector; compare sentences with cosine similarity. Team (Universidad Politécnica de Yucatán, U2T02): Acosta Castellanos Bianca Alexandra, Canche Chuc Angel Rivaldo, Ku Aguilar Russel, Sanchez Novelo Damian, Velasco Martin Jonathan Abisai. 165,528 unique sentences (premises and hypotheses, whitespace-stripped) from snlitrain100k.jsonl, a 100k-record subset of the SNLI train split. Labels are not used: each sentence is its own positive, the two views differ only by dropout. The paper…

static quants of https://huggingface.co/reasonwang/SkillGym-Qwen3.5-9B 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 is a reranker fine-tuned for AI agent skill retrieval. Given a natural-language user request and a candidate agent skill, it scores how relevant and useful the skill is for the request. It is designed as the second stage after a first-stage retriever such as SkillRet-Embedding-0.6B or SkillRet-Embedding-8B. The model is fine-tuned from Qwen/Qwen3-Reranker-0.6B on the SkillRet benchmark training split with binary cross-entropy on the yes/no token probability. It keeps the scoring interface of Qwen3-Reranker. Each skill document is body, the same representation used by the SkillRet embedding models. Evaluated on the SkillRet benchmark evaluation split (4,392 queries, 6,006 skills). The…

EfficientNetV2-L (torchvision efficientnetv2l, 1000 ImageNet classes) fine-tuned with adversarial training for the Perturb subnet (netuid 26).

Model · Text generation

Slayer149-balanced

Fabryka AI

Experimental 149,333,081-parameter English causal model trained from scratch. 26 layers, hidden width 576, FFN 2496, 9 query heads / 3 KV heads, tied 24,576-token byte-level BPE vocabulary, QK normalization and value residuals. Architecture informed by Qwen3 and AltSlate's Apache-2.0 Jugnu recipe; no Jugnu weights reused. This is a separate experiment from SlayerLab/Slayer149, which remains available. Training may still be in progress. See reports/status.json, training metrics and checkpoint manifests. No leaderboard rank or winning result is claimed. The final full GLINT-derived evaluation is reports/evaluation.json, when present. Competitor scores may use different evaluation conventions.…

SlimSAM (nielsr/slimsam-77-uniform) in ONNX for transformers.js, like Xenova/slimsam-77-uniform — whose image encoder files are reused here unchanged — but with a prompt encoder / mask decoder that takes box prompts. Xenova/slimsam-77-uniform's decoder declares no inputboxes, and transformers.js drops inputs a session does not declare, so a box passed to it is silently ignored. This decoder follows Meta's own SAM ONNX convention instead: a box is two extra points in inputpoints, labelled 2 (top-left corner) and 3 (bottom-right corner). Inputs and outputs keep transformers.js's names, so SamModel / SamProcessor drive it unchanged. One rule the caller must follow: a prompt without a box…

1,427,969,472 trainable parameters, initialized from pinned Wan2.1-T2V-1.3B. Frozen native video codec, 512×288 at 15 Hz, three seconds observed history, 30 Hz proxy angle/boost controls. Joint stochastic future blocks with native sigma flow matching. Policy fitting is a later separate stage. Resumable latest and best checkpoints: https://huggingface.co/buckets/mundoamundo/slither-wan-v3-20261001. Read latest.json or best.json for verified file pointers, sizes, and SHA-256. Two rotating slots per kind; no repeated weight commits to this Git repository. Checkpoints include model, optimizer, data cursor, and per-rank RNG. Dataset: mundoamundo/slither-wam-video-actions at…

These are weight-only FP8 kernels for NVIDIA Volta (V100, SM70). They serve checkpoints whose weights are E4M3 with one power-of-two (UE8M0) scale per 32 × 32 block, the dense format of DeepSeek-V4.1-Flash (weightblocksize: [32, 32], scalefmt: ue8m0). Volta has no FP8 tensor cores, so the weights stay FP8 in memory and are expanded to FP16 inside the kernels. Weight memory falls from 2 bytes per parameter (FP16 dequantization) to 1.0625 bytes. For DeepSeek-V4.1-Flash's per-layer dense set at TP4, decode-size products (1–16 rows) run 9–36% faster than FP16 weights with cuBLAS. Individual projections vary (see Results). This repository is a distribution of a patch series for 1Cat-vLLM, the…

Extracts fifteen figures from the primary financial statements of a US SEC 10-K or 10-Q, as JSON. A LoRA fine-tune of google/gemma-4-E2B at revision d29ff6b45f081a49ee2733a859c9c9c2d95d1a6f, trained on filings with their XBRL facts as the labels. No model graded another model anywhere in producing these numbers. 705 filings. On filings published before the base model's training cutoff it scores 95.2% (94.7% to 95.5%), against 96.9% (96.5% to 97.3%) after it. That difference is the contamination gap, reported and never used as the headline. Non-inferiority field by field, paired on the same filings, a three-point margin per field and Holm's adjustment across the fifteen, decided by the…

Extracts fifteen figures from the primary financial statements of a US SEC 10-K or 10-Q, as JSON. A LoRA fine-tune of google/gemma-4-E2B at revision d29ff6b45f081a49ee2733a859c9c9c2d95d1a6f, trained on filings with their XBRL facts as the labels. No model graded another model anywhere in producing these numbers. 705 filings. On filings published before the base model's training cutoff it scores 95.2% (94.7% to 95.5%), against 96.9% (96.5% to 97.3%) after it. That difference is the contamination gap, reported and never used as the headline. Non-inferiority field by field, paired on the same filings, a three-point margin per field and Holm's adjustment across the fifteen, decided by the…

static quants of https://huggingface.co/Peter-A-P/smallprint-2b 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 quant And…

Extracts fifteen figures from the primary financial statements of a US SEC 10-K or 10-Q, as JSON. A LoRA fine-tune of google/gemma-4-E2B at revision d29ff6b45f081a49ee2733a859c9c9c2d95d1a6f, trained on filings with their XBRL facts as the labels. No model graded another model anywhere in producing these numbers. 705 filings. On filings published before the base model's training cutoff it scores 95.2% (94.7% to 95.5%), against 96.9% (96.5% to 97.3%) after it. That difference is the contamination gap, reported and never used as the headline. Non-inferiority field by field, paired on the same filings, a three-point margin per field and Holm's adjustment across the fifteen, decided by the…

Extracts fifteen figures from the primary financial statements of a US SEC 10-K or 10-Q, as JSON. A LoRA fine-tune of google/gemma-4-E4B at revision 411aa17b749aa952df1359d2dcea73917a544d9a, trained on filings with their XBRL facts as the labels. No model graded another model anywhere in producing these numbers. 705 filings. On filings published before the base model's training cutoff it scores 95.1% (94.7% to 95.4%), against 96.9% (96.5% to 97.3%) after it. That difference is the contamination gap, reported and never used as the headline. Non-inferiority field by field, paired on the same filings, a three-point margin per field and Holm's adjustment across the fifteen, decided by the…

Extracts fifteen figures from the primary financial statements of a US SEC 10-K or 10-Q, as JSON. A LoRA fine-tune of google/gemma-4-E4B at revision 411aa17b749aa952df1359d2dcea73917a544d9a, trained on filings with their XBRL facts as the labels. No model graded another model anywhere in producing these numbers. 705 filings. On filings published before the base model's training cutoff it scores 95.1% (94.7% to 95.4%), against 96.9% (96.5% to 97.3%) after it. That difference is the contamination gap, reported and never used as the headline. Non-inferiority field by field, paired on the same filings, a three-point margin per field and Holm's adjustment across the fifteen, decided by the…