Open-Weight Models

Model

ocr-freeform

Liam Foster

This repository contains a working research note about OCR Freeform. It organizes motivation, related work, a falsifiable hypothesis, and an evaluation plan. It is not presented as a completed paper or a release of trained models. - the scope of the research question and likely confounders - a proposed comparison with matched baselines - concrete evaluation context such as FUNSD, SROIE, and CORD - reproducibility checks, failure modes, and open questions - topic-relevant references Start with summary.md for the full note. Sections labeled as plans or hypotheses should not be interpreted as experimental results. If results are added later, they should include dataset versions, commands…

Model

ocr-freeform-test

Jungkookjeon1994

This repository contains a working research note about OCR Freeform. It organizes motivation, related work, a falsifiable hypothesis, and an evaluation plan. It is not presented as a completed paper or a release of trained models. - the scope of the research question and likely confounders - a proposed comparison with matched baselines - concrete evaluation context such as FUNSD, SROIE, and CORD - reproducibility checks, failure modes, and open questions - topic-relevant references Start with analysis.md for the full note. Sections labeled as plans or hypotheses should not be interpreted as experimental results. If results are added later, they should include dataset versions, commands…

Model

ocr-freeform-v2

N. Ramirez

This is an exploratory note for OCR Freeform. It records the intended comparison, likely confounders, and reproducibility requirements before any benchmark result is reported. - the scope of the research question and likely confounders - a proposed comparison with matched baselines - concrete evaluation context such as FUNSD, SROIE, and CORD - reproducibility checks, failure modes, and open questions - topic-relevant references Start with analysis.md for the full note. Sections labeled as plans or hypotheses should not be interpreted as experimental results. If results are added later, they should include dataset versions, commands, seeds, hardware, and raw logs. The note is intentionally…

Fine-tuned omniASRLLM1Bv2 (run 3, step 20,000) blended with the base model at α = 0.85 (WiSE-FT weight interpolation). Use this model for monolingual Indian-language speech. English, Hinglish / code-mixed speech and all other languages are served by v1 (same run, α = 0.5). The α = 0.7 version of this file is in the repo history. α 1.0 was rejected: FLEURS dev (63 languages) 13.63 vs 13.51 for α 0.5, 38 languages worse, and the worst Hinglish / Indian-English scores. Paired bootstrap, α 0.85 vs α 0.7 on IndicVoices dev: −2.00 [−3.12, −1.15], P(better) = 1.00. \ Same-domain: training included IndicVoices train audio. Full α 0.85 test halves and FLEURS-Indic are scored after the current GPU…

OD-1 is an open System One decision model: give it a state (text or JSON) and typed questions (choice, noul = yes/no, score = ordinal) and it returns every answer from one forward pass, with full probability distributions including an explicit NOTANSWERABLE outcome. The request/response shape follows the Jev / TypeSafe /v1/systemone wire format. Backbone: Qwen/Qwen3.5-0.8B (Apache-2.0) with typed answer heads read at marker tokens, early-exit heads (layers 6/12/18), and a CLM-style contrastive shortlist head for questions with more than 24 options. Trained with SFT + distillation from Qwen3.5-27B-FP8 (never a compared system), RLCD-style proper-score calibration, an abstention utility…

OD-1 is an open System One decision model: give it a state (text or JSON) and typed questions (choice, noul = yes/no, score = ordinal) and it returns every answer from one forward pass, with full probability distributions including an explicit NOTANSWERABLE outcome. The request/response shape follows the Jev / TypeSafe /v1/systemone wire format. Backbone: Qwen/Qwen3.5-0.8B (Apache-2.0) with typed answer heads read at marker tokens, early-exit heads (layers 6/12/18), and a CLM-style contrastive shortlist head for questions with more than 24 options. Trained with SFT + distillation from Qwen3.5-27B-FP8 (never a compared system), RLCD-style proper-score calibration, an abstention utility…

OD-1 is an open System One decision model: give it a state (text or JSON) and typed questions (choice, noul = yes/no, score = ordinal) and it returns every answer from one forward pass, with full probability distributions including an explicit NOTANSWERABLE outcome. The request/response shape follows the Jev / TypeSafe /v1/systemone wire format. Backbone: Qwen/Qwen3.5-0.8B (Apache-2.0) with typed answer heads read at marker tokens, early-exit heads (layers 6/12/18), and a CLM-style contrastive shortlist head for questions with more than 24 options. Trained with SFT + distillation from Qwen3.5-27B-FP8 (never a compared system), RLCD-style proper-score calibration, an abstention utility…

One street photo in, a structured Japanese observation record out. OddEye is the street-reading model of michiyomi. It is built on Qwen3.8-27B and compressed to 4-bit (NVFP4, about 20 GB), so a single GPU — even an RTX 5090 — can run it. It keeps what it sees (observations) and what it infers (interpretations, each with a confidence label) in separate fields. The record has 13 top-level fields, including objects, text on signs, relations, surfaces, standard road facets (lanes, sidewalks, lighting…), interpretations and a scene summary. All numbers use the recommended prompt (v5.2c), greedy decoding and the same photos for every model. "Generic NVFP4" is a publicly distributed 4-bit version…

Model

odyssey-models

ADRLAB

Checkpoints of the end-to-end driving planners evaluated in the Odyssey closed-loop benchmark. - baseline: the upstream model. It takes NAVSIM's drivingcommand and has no route. - sdroute: drivingcommand is removed, and a standard-definition (SD) route becomes the only navigation signal. - The code that loads and runs these checkpoints is OdysseyZoo. A link will be added here once it is public. - The closed-loop scenes are in ADRLAB/odyssey-scenes. - The simulator's image restorer (Fixer, fine-tuned from NVIDIA Fixer) is in fixer/. See Fixer. Paths match the OdysseyZoo tree, so the files land in place when you download into an OdysseyZoo checkout. - Every planner checkpoint was trained on…

Complete code and saved checkpoints for Square and ToolHang, seed 1, comparing current-actor confidence ordering with random position ordering. Success buffer is disabled in both selected variants. No license grant is implied by this export. - checkpoints/ / /: saved policies, identical per-task BC anchors, latest full replay buffer and rolling metadata, metric CSVs, and model weights. - analysisinputs/: a fixed 4096-observation sample per task from the earlier UMAP snapshot (2048 per variant, seed 42), for shared-input representation comparisons. Only optimizer-state dictionaries (optstate) are removed from model checkpoints. Current actor, target actor, current critic, target critic…

Engineered & Architected by Suryaansh Prithvijit Singh (@Ultron28 / @Ultron09) The ULTRON Financial Conformer (1.189B) is a large-scale foundation model specifically designed to ingest raw minute-level OHLCV candlestick time series across liquid US equities and generate multi-horizon continuous returns and directional probability forecasts. Unlike conventional time-series transformers (e.g. PatchTST, Chronos, TimesFM) that treat financial candlesticks as 1D scalar sequences, this architecture treats the microstructure as a continuous multi-scale physical manifold. It combines: 1. 6-Scale Temporal Inception Tokenizer: Continuous 1D convolutions spanning $[3\text{m}, 7\text{m}, 15\text{m}…

PaddlePaddle's PP-FormulaNetplus-L, which reads a picture of a formula as LaTeX, converted to Core ML ML programs with 8-bit weights for OhMyShot, a screenshot app for macOS, which downloads them as a model pack to copy the formula in a selection as LaTeX. The weights are PaddlePaddle's, quantized; see LICENSE.txt for the license and what was changed. Decoding starts from token 2 at position 0 and ends when the decoder gives 2. Run on the CPU (.cpuOnly). Made with Tools/model-packs/formulapack.py from this upstream revision: - https://huggingface.co/PaddlePaddle/PP-FormulaNetplus-Lsafetensors/tree/6cc004a8e91ea3f3819b05317f40b290cbe99c2b Checked against the OhMyShot formula corpus…

The model weights of the LoRA / QLoRA runs done on the L40S server during the Warsaw Model Trainers Hackathon (25-26.09.2026): PEFT adapters, Trainer checkpoints (with optimizer state), lora.gguf adapters for llama.cpp and trainingargs.bin. These are adapters, not full models. Every run uses the base google/gemma-4-12B-it-qat-q40-unquantized (the local path in adapterconfig.json points to a server copy of it). Layout: l40s-results/ /..., the same paths as oldwb/l40s-results/ in the GitHub repo Endote/WarsawModelTrainersHackathon, which holds the logs, metrics, data statistics and write-ups (README.md, EXPERIMENTS.md, BENCHMARKSRESEARCH.md). Restore into a checkout of that repo: python…

Every arm here differs from the recipe of record by exactly one flag. Data, pool, crop seed, layer set, learning rate, batch and seeds are held fixed across arms so the only free variable is the one named. The whitened-FVE ruler that the program optimises and reports anti-correlates with lens quality. The AR trained without whitening scores far worse on the ruler and makes a materially better lens, and running both to saturation does not rescue it. Compare arms only at MATCHED cumulative steps. Arms stopped early by a budget cut have lower totals, and reading their endpoints against a fully-trained arm inverts the ordering: unitcos looks worse than base on endpoints but is identical to it…

基于 MikeLee000/olmo3-190m-zh-full 完成一轮持续预训练（Continued Pretraining，CPT）的课程实验模型，约 187M 参数。 不是 SFT 或指令对话模型。 使用普通文本作为提示进行续写，不使用 applychattemplate，不应将其作为可靠的问答助手。 - 架构：Olmo3ForCausalLM，12 层，hidden size 768，intermediate size 3072，词表 48,000。 - 数据：cmz1024/llm101-olmo3-zh-demo-data 仓库中的 continue/tokenizedv31.bin。 - 全参数持续预训练；不使用 LoRA；序列长度 2,048；BF16。 - 1 epoch，共 11,918 个 optimizer steps。 - Batch size 16，gradient accumulation 4，有效 batch size 64。 - Learning rate 2e-4，warmup ratio 0.1，cosine scheduler。 - Transformers 5.9.0，PyTorch 2.5.1。 - 最终 validation loss：1.9467759132385254。这是特定验证集上的语言模型损失，不是准确率或通用问答质量评分。 发布文件来自 /checkpoints/olmo3zhcontinue/final，已核对与 checkpoint-11918/model.safetensors 的 SHA-256 一致：…