suayptalha

This project performs full fine-tuning on the Qwen3-0.6B language model to enhance its mathematical problem-solving and reasoning capabilities. Training was conducted exclusively on the `OpenMathReasoning-mini` dataset, and the model was optimized using the bfloat16 (bf16) data type. The `unsloth/OpenMathReasoning-mini` dataset was used. Each example was formatted in Chain-of-Thought (CoT) style, pairing math problems with step-by-step intermediate reasoning. Qwen3 base model weights were loaded via the `unsloth` library in bf16 precision. All layers were updated (`fullfinetuning=True`) to adapt the model for mathematical reasoning. Leveraged the Hugging Face TRL library with the Supervised Fine-Tuning (SFT) approach. The model was trained to generate both correct answers and corresponding reasoning chains. The model’s reasoning capacity for math problems was significantly improved through single-dataset, full fine-tuning in bf16 precision. Outputs include both intermediate reasoning steps and final solutions, providing transparent and interpretable results. This project is licensed under the Apache License 2.0. See the LICENSE file for details.

This is the quantized version of suayptalha/Sungur-9B. Sungur-9B is a Turkish-specialized large language model derived from ytu-ce-cosmos/Turkish-Gemma-9b-v0.1, which itself is based on Gemma-2-9b. The model was further trained using a 7k-sample Direct Preference Optimization (DPO) dataset created via translation and fine-tuned with 4-bit QLoRA, refining its alignment with human preferences. Sungur-9B is designed for Turkish text generation tasks, producing coherent and contextually appropriate outputs. Its training process enables it to deliver fluent, context-aware responses. Turkish Evaluation Benchmark Results (via `malhajar17/lm-evaluation-harnessturkish`) | Task / Dataset | suayptalha/Sungur-9B | Qwen/Qwen2.5-7B-Instruct | google/gemma-2-9b-it | ytu-ce-cosmos/Turkish-Gemma-9b-v0.1 | google/gemma-3-12b-it | Qwen/Qwen2.5-14B-it | Qwen/Qwen2.5-32B-Instruct | google/gemma-2-27b-it | google/gemma-3-27b-it | Qwen/Qwen2.5-72B-Instruct | meta-llama/Llama-3-1-70B-Instruct | | -------------------- | ------------------------ | ------------------------ | -------------------- | ----------------------------------- | --------------------- | ------------------- | ------------------------- | --------------------- | --------------------- | ------------------------- | --------------------------------- | | MMLU (tr) | 61.19 | 56.31 | 61.07 | 63.85 | 63.92 | 65.28 | 70.93 | 66.49 | 70.20 | 77.28 | 74.00 | | TruthfulQA (tr) | 55.21 | 55.99 | 55.77 | 54.21 | 57.16 | 59.00 | 57.87 | 57.45 | 57.06 | 59.86 | 51.41 | | ARC (tr) | 55.03 | 42.06 | 56.31 | 59.64 | 60.67 | 50.00 | 57.00 | 63.65 | 66.98 | 61.52 | 59.64 | | Hellaswag (tr) | 64.36 | 44.71 | 56.48 | 64.19 | 62.00 | 52.22 | 57.04 | 63.86 | 66.58 | 61.98 | 64.31 | | Gsm8K (tr) | 74.49 | 64.16 | 63.10 | 73.42 | 72.06 | 76.77 | 77.83 | 76.54 | 77.52 | 83.60 | 66.13 | | Winogrande (tr) | 63.43 | 59.66 | 62.09 | 64.53 | 61.77 | 58.77 | 61.77 | 65.40 | 65.80 | 61.92 | 66.90 | Acknowledgments - Thanks to ytu-ce-cosmos for their amazing Turkish-Gemma-9b-v0.1 model. - Thanks to axolotl for making the repository I used to make this model. - Thanks to all Turkish open source AI community.

This project performs full fine-tuning on the Qwen3-0.6B language model to enhance its medical reasoning and clinical understanding capabilities. Training was conducted on the `FreedomIntelligence/medical-o1-reasoning-SFT` dataset using bfloat16 (bf16) precision for efficient optimization. The `FreedomIntelligence/medical-o1-reasoning-SFT` dataset was used. Each example consists of medically relevant instructions or questions paired with detailed, step-by-step clinical reasoning responses. Prompts were structured to encourage safe, factual, and coherent medical reasoning chains. Qwen3 base model weights were loaded via the `unsloth` library in bf16 precision. All model layers were fully updated (`fullfinetuning=True`) to effectively adapt the model to medical reasoning and decision-making tasks. Fine-tuning was conducted using the Hugging Face TRL library with the Supervised Fine-Tuning (SFT) approach. The model was trained to follow clinical instructions, interpret symptoms, and generate reasoned diagnoses or treatment suggestions. The model’s ability to interpret medical instructions and generate step-by-step clinical reasoning has been significantly enhanced. It produces responses that combine factual accuracy with transparent reasoning, making it useful in educational and assistive medical AI contexts. This project is licensed under the Apache License 2.0. See the LICENSE file for details.

Sungur-9B is a Turkish-specialized large language model derived from ytu-ce-cosmos/Turkish-Gemma-9b-v0.1, which itself is based on Gemma-2-9b. The model was further trained using a 7k-sample Direct Preference Optimization (DPO) dataset created via translation and fine-tuned with 4-bit QLoRA, refining its alignment with human preferences. Sungur-9B is designed for Turkish text generation tasks, producing coherent and contextually appropriate outputs. Its training process enables it to deliver fluent, context-aware responses. Turkish Evaluation Benchmark Results (via `malhajar17/lm-evaluation-harnessturkish`) | Task / Dataset | suayptalha/Sungur-9B | Qwen/Qwen2.5-7B-Instruct | google/gemma-2-9b-it | ytu-ce-cosmos/Turkish-Gemma-9b-v0.1 | google/gemma-3-12b-it | Qwen/Qwen2.5-14B-it | Qwen/Qwen2.5-32B-Instruct | google/gemma-2-27b-it | google/gemma-3-27b-it | Qwen/Qwen2.5-72B-Instruct | meta-llama/Llama-3-1-70B-Instruct | | -------------------- | ------------------------ | ------------------------ | -------------------- | ----------------------------------- | --------------------- | ------------------- | ------------------------- | --------------------- | --------------------- | ------------------------- | --------------------------------- | | MMLU (tr) | 61.19 | 56.31 | 61.07 | 63.85 | 63.92 | 65.28 | 70.93 | 66.49 | 70.20 | 77.28 | 74.00 | | TruthfulQA (tr) | 55.21 | 55.99 | 55.77 | 54.21 | 57.16 | 59.00 | 57.87 | 57.45 | 57.06 | 59.86 | 51.41 | | ARC (tr) | 55.03 | 42.06 | 56.31 | 59.64 | 60.67 | 50.00 | 57.00 | 63.65 | 66.98 | 61.52 | 59.64 | | Hellaswag (tr) | 64.36 | 44.71 | 56.48 | 64.19 | 62.00 | 52.22 | 57.04 | 63.86 | 66.58 | 61.98 | 64.31 | | Gsm8K (tr) | 74.49 | 64.16 | 63.10 | 73.42 | 72.06 | 76.77 | 77.83 | 76.54 | 77.52 | 83.60 | 66.13 | | Winogrande (tr) | 63.43 | 59.66 | 62.09 | 64.53 | 61.77 | 58.77 | 61.77 | 65.40 | 65.80 | 61.92 | 66.90 | Acknowledgments - Thanks to ytu-ce-cosmos for their amazing Turkish-Gemma-9b-v0.1 model. - Thanks to axolotl for making the repository I used to make this model. - Thanks to all Turkish open source AI community.

This project performs full fine-tuning on the Qwen3-0.6B language model to enhance its code reasoning and generation capabilities. Training was conducted exclusively on the `nvidia/OpenCodeReasoning` dataset, and the model was optimized using the bfloat16 (bf16) data type. `nvidia/OpenCodeReasoning` dataset was used. Each example consists of code snippets paired with detailed step-by-step reasoning in Chain-of-Thought (CoT) style. Qwen3-0.6B base model weights were loaded via the `unsloth` library in bf16 precision. Full fine-tuning (`fullfinetuning=True`) was applied to all layers for optimal adaptation to code reasoning. Employed the Hugging Face TRL library with the Supervised Fine-Tuning (SFT) approach. The model was trained to generate correct code solutions along with the corresponding reasoning chains. The model’s capacity for understanding, reasoning about, and generating code was significantly improved through specialized, single-dataset training in bf16 precision. Outputs include both intermediate reasoning steps and final code solutions, enabling transparent and interpretable code generation. This project is licensed under the Apache License 2.0. See the LICENSE file for details.

This is the quantized version of suayptalha/Sungur-14B. Sungur-14B is a Turkish-specialized large language model derived from Qwen/Qwen3-14B. The model was fine-tuned using suayptalha/Sungur-Dataset, a 41.1k-sample collection of reasoning-focused conversations spanning domains such as mathematics, medicine, and general knowledge. This dataset is entirely in Turkish and was created to enhance native Turkish reasoning ability. The training process employed 4-bit QLoRA for Supervised Fine-Tuning (SFT), enabling efficient adaptation while preserving the capabilities of the base model. Sungur-14B is designed for Turkish reasoning and text generation tasks, delivering coherent, context-aware, and logically structured responses. Through its specialized dataset and training pipeline, the model gains strong native reasoning capabilities in Turkish, making it suitable for advanced applications in analytical dialogue, education, and domain-specific problem solving. > [!NOTE] > For thinking mode, use `temperature=0.6`, `topp=0.95`, `topk=20`, `minp=0`, and `repetitionpenalty=1.2`. DO NOT use greedy decoding, as it can lead to performance degradation and endless repetitions. > For non-thinking mode, use `temperature=0.7`, `topp=0.8`, `topk=20`, and `minp=0`. This model was trained on `1xB200` GPU. Training took ~3 hours. Comparison with Base Model (via `malhajar17/lm-evaluation-harnessturkish`) | Benchmark | Sungur-14B | Qwen3-14B | | ------------------------ | ---------- | ---------- | | ARC (tr, acc) | 0.4727 | 0.4701 | | ARC (tr, accnorm) | 0.5213 | 0.5273 | | GSM8K (tr, flex) | 0.0380 | 0.0418 | | GSM8K (tr, strict) | 0.7760 | 0.8185 | | HellaSwag (tr, acc) | 0.4051 | 0.4017 | | HellaSwag (tr, norm) | 0.5279 | 0.5113 | | Winogrande (tr) | 0.5893 | 0.5656 | | TruthfulQA (acc) | 0.5174 | 0.5165 | | MMLU (tr, ort.) | 0.6640 | 0.6729 | | Model Name | GSM8K (strict) | | --------------------------------------- | -------------- | | Qwen/Qwen2.5-72B-Instruct | 83.60 | | Qwen/Qwen3-14B | 81.85 | | Qwen/Qwen2.5-32B-Instruct | 77.83 | | suayptalha/Sungur-14B | 77.60 | | google/gemma-3-27b-it | 77.52 | | ytu-ce-cosmos/Turkish-Gemma-9b-T1 | 77.41 | | Qwen/Qwen2.5-14B-it | 76.77 | | google/gemma-2-27b-it | 76.54 | | suayptalha/Sungur-9B | 74.49 | | ytu-ce-cosmos/Turkish-Gemma-9b-v0.1 | 73.42 | | google/gemma-3-12b-it | 72.06 | | meta-llama/Llama-3-1-70B-Instruct | 66.13 | | Qwen/Qwen2.5-7B-Instruct | 64.16 | | google/gemma-2-9b-it | 63.10 | Acknowledgments - Thanks to @Qwen team for their amazing Qwen/Qwen3-14B model. - Thanks to unsloth for making the repository I used to make this model. - Thanks to all Turkish open source AI community.

This project performs full fine-tuning on the Qwen3-0.6B language model to enhance its instruction-following and reasoning capabilities. Training was conducted on the `patrickfleith/instruction-freak-reasoning` dataset using bfloat16 (bf16) precision for efficient optimization. The `patrickfleith/instruction-freak-reasoning` dataset was used. Each example contains a complex instruction paired with an in-depth reasoning-based response. Prompts were structured to encourage chain-of-thought style outputs when applicable. Qwen3 base model weights were loaded via the `unsloth` library in bf16 precision. All model layers were fully updated (`fullfinetuning=True`) to effectively adapt the model to instruction understanding and stepwise response generation. Fine-tuning was conducted using the Hugging Face TRL library with the Supervised Fine-Tuning (SFT) approach. The model was trained to follow detailed instructions, reason logically, and generate structured responses. The model’s ability to follow complex instructions and explain its reasoning process has been significantly enhanced. It generates both coherent reasoning steps and conclusive answers, improving transparency and usability for instruction-based tasks. This project is licensed under the Apache License 2.0. See the LICENSE file for details.

Anime-Otaku-Qwen-Image is a LoRA fine-tuned adapter for the Qwen-Image model, trained on the Anime Gen v2 dataset. It specializes in generating high-quality anime-style images. Fine-tuned on xingjianleng/animegenv2 dataset. Generates anime characters, scenes, and illustrations in a consistent style. LoRA adapter (\~rank 16) — lightweight and easy to use with the base Qwen-Image model. Compatible with Hugging Face Diffusers pipelines. > Make sure your prompt starts with "Anime, " to match the training data format. Recommended inference resolution: 1024×1024. You can reduce to 512×512 if GPU memory is limited. Use `torchdtype=torch.bfloat16` or `torch.float16` for efficient VRAM usage. Enable attention slicing to reduce memory usage: You should use `Anime` to trigger the image generation.

Sungur-14B is a Turkish-specialized large language model derived from Qwen/Qwen3-14B. The model was fine-tuned using suayptalha/Sungur-Dataset, a 41.1k-sample collection of reasoning-focused conversations spanning domains such as mathematics, medicine, and general knowledge. This dataset is entirely in Turkish and was created to enhance native Turkish reasoning ability. The training process employed 4-bit QLoRA for Supervised Fine-Tuning (SFT), enabling efficient adaptation while preserving the capabilities of the base model. Sungur-14B is designed for Turkish reasoning and text generation tasks, delivering coherent, context-aware, and logically structured responses. Through its specialized dataset and training pipeline, the model gains strong native reasoning capabilities in Turkish, making it suitable for advanced applications in analytical dialogue, education, and domain-specific problem solving. This model was trained on `1xB200` GPU. Training took ~3 hours. By default, Sungur-14B has thinking capabilities enabled. This means the model will use its reasoning abilities to enhance the quality of generated responses. For example, when explicitly setting `enablethinking=True` or leaving it as the default value in `tokenizer.applychattemplate`, the model will engage its thinking mode. If you do not want thinking you can set `enablethinking=False`. > [!NOTE] > For thinking mode, use `temperature=0.6`, `topp=0.95`, `topk=20`, `minp=0`, and `repetitionpenalty=1.2`. DO NOT use greedy decoding, as it can lead to performance degradation and endless repetitions. > For non-thinking mode, use `temperature=0.7`, `topp=0.8`, `topk=20`, and `minp=0`. Comparison with Base Model (via `malhajar17/lm-evaluation-harnessturkish`) | Benchmark | Sungur-14B | Qwen3-14B | | ------------------------ | ---------- | ---------- | | ARC (tr, acc) | 0.4727 | 0.4701 | | ARC (tr, accnorm) | 0.5213 | 0.5273 | | GSM8K (tr, flex) | 0.0380 | 0.0418 | | GSM8K (tr, strict) | 0.7760 | 0.8185 | | HellaSwag (tr, acc) | 0.4051 | 0.4017 | | HellaSwag (tr, norm) | 0.5279 | 0.5113 | | Winogrande (tr) | 0.5893 | 0.5656 | | TruthfulQA (acc) | 0.5174 | 0.5165 | | MMLU (tr, ort.) | 0.6640 | 0.6729 | | Model Name | GSM8K (strict) | | --------------------------------------- | -------------- | | Qwen/Qwen2.5-72B-Instruct | 83.60 | | Qwen/Qwen3-14B | 81.85 | | Qwen/Qwen2.5-32B-Instruct | 77.83 | | suayptalha/Sungur-14B | 77.60 | | google/gemma-3-27b-it | 77.52 | | ytu-ce-cosmos/Turkish-Gemma-9b-T1 | 77.41 | | Qwen/Qwen2.5-14B-it | 76.77 | | google/gemma-2-27b-it | 76.54 | | suayptalha/Sungur-9B | 74.49 | | ytu-ce-cosmos/Turkish-Gemma-9b-v0.1 | 73.42 | | google/gemma-3-12b-it | 72.06 | | meta-llama/Llama-3-1-70B-Instruct | 66.13 | | Qwen/Qwen2.5-7B-Instruct | 64.16 | | google/gemma-2-9b-it | 63.10 | Acknowledgments - Thanks to @Qwen team for their amazing Qwen/Qwen3-14B model. - Thanks to unsloth for making the repository I used to make this model. - Thanks to all Turkish open source AI community.

This is a English to Turkish translator t5-base finetuned model on Helsinki-NLP/opus-100 License: This model is on Apache-2.0 License. Check licence docs for more!

First Hugging Face integration of minGRU models from the paper "Were RNNs All We Needed?". This model uses BERT-Base-Uncased tokenizer and trained on default IMDB dataset. Make sure you have installed "minGRU-pytorch" library by running "pip install minGRU-pytorch". > Text: The movie was absolutely wonderful, I loved it! > Predicted sentiment: positive I am thankful to Leo Feng, Frederick Tung, Mohamed Osama Ahmed, Yoshua Bengio and Hossein Hajimirsadeghi for their papers.

"We are all experts at something, but we’re all also beginners at something else." This is a MoE (Mixture of Experts) Qwen3 model which has total number of 2.4B parameters and 0.6B for each of 4 experts. All the expert models can be seen below. This model aims to provide more accurate results with more efficiency and less memory usage! `perdevicetrainbatchsize = 2` `gradientaccumulationsteps = 4` `warmupsteps = 5` `numtrainepochs = 1` `learningrate = 2e-5` `optim = "adamw8bit"` `weightdecay = 0.01` `seed = 3407` This model was fully fine-tuned with BF16 on first 20k rows of `nvidia/OpenCodeReasoning` dataset for 1 epoch. This model was fully fine-tuned with BF16 on entire `unsloth/OpenMathReasoning-mini` dataset for 1 epoch. This model was fully fine-tuned with BF16 on first 20k rows of `FreedomIntelligence/medical-o1-reasoning-SFT` dataset for 1 epoch. `Qwen/Qwen3-0.6B` model was directly used for this expert, no fine-tune was applied. Router Model: The router model can be found here which was trained version of `distilbert/distilbert-base-uncased` on 7 different datasets. This project is licensed under the Apache License 2.0. See the LICENSE file for details.

ebody { font-family: 'Quicksand', sans-serif; background: linear-gradient(135deg, #FF69B4 0%, #800080 100%); color: #FFFFFF; margin: 0; padding: 0; font-size: 16px; min-height: 100vh; } .container { margin: 20px; background-color: rgba(28, 14, 36, 0.95); padding: 20px; border-radius: 12px; box-shadow: 0 4px 20px rgba(255, 105, 180, 0.4); border: 1px solid rgba(255, 105, 180, 0.4); outline: 1px solid rgba(255, 105, 180, 0.7); outline-offset: -1px; position: relative; backdrop-filter: blur(10px); } .container::before { content: ''; position: absolute; top: -1px; left: -1px; right: -1px; bottom: -1px; border: 1px solid rgba(255, 105, 180, 0.98); border-radius: 12px; pointer-events: none; animation: borderGlow 2s ease-in-out infinite; } @keyframes borderGlow { 0% { box-shadow: 0 0 5px rgba(255, 105, 180, 0.98); } 50% { box-shadow: 0 0 20px rgba(255, 105, 180, 0.98); } 100% { box-shadow: 0 0 5px rgba(255, 105, 180, 0.98); } } .header h1 { font-size: 28px; color: #FF69B4; margin: 0 0 20px 0; text-shadow: 0 0 15px rgba(255, 105, 180, 0.8); letter-spacing: 1px; } .update-section h2, h2 { font-size: 24px; color: #FF69B4; text-shadow: 0 0 15px rgba(255, 105, 180, 0.8); letter-spacing: 0.5px; } .update-section p { font-size: 16px; line-height: 1.6; color: #FFE1FF; } .info p { color: #FFE1FF; line-height: 1.6; font-size: 16px; } .info img { width: 100%; border-radius: 10px; margin-bottom: 15px; box-shadow: 0 0 30px rgba(255, 105, 180, 0.5); border: 1px solid rgba(255, 105, 180, 0.4); outline: 1px solid rgba(255, 105, 180, 0.7); outline-offset: -1px; transition: transform 0.3s ease, box-shadow 0.3s ease; } .info img:hover { transform: scale(1.01); box-shadow: 0 0 40px rgba(255, 105, 180, 0.6); } a { color: #00FFEE; text-decoration: none; transition: color 0.3s ease; } .button { display: inline-block; background: linear-gradient(45deg, rgba(255, 105, 180, 0.9), rgba(128, 0, 128, 0.9)); color: #FFFFFF; padding: 12px 24px; border-radius: 5px; cursor: pointer; text-decoration: none; transition: all 0.3s ease; border: 1px solid rgba(255, 105, 180, 0.4); } .button:hover { background: linear-gradient(45deg, rgba(255, 105, 180, 1), rgba(128, 0, 128, 1)); box-shadow: 0 0 20px rgba(255, 105, 180, 0.7); transform: translateY(-2px); } pre { background-color: rgba(28, 14, 36, 0.95); padding: 15px; border-radius: 5px; overflow-x: auto; border: 1px solid rgba(255, 20, 147, 0.3); outline: 1px solid rgba(255, 20, 147, 0.6); outline-offset: -1px; } code { font-family: 'Courier New', monospace; color: #FFE1FF; } .info-grid { display: grid; grid-template-columns: repeat(3, 1fr); gap: 15px; } .creator-label { color: #FFE1FF; font-size: 14px; margin-right: 8px; } .creator-link { display: flex; align-items: center; gap: 5px; color: #00FFEE; text-decoration: none; transition: all 0.3s ease; } .creator-arrow { font-size: 16px; transition: transform 0.3s ease; } .creator-link:hover .creator-arrow { transform: translateX(3px); } .name-legend { background: rgba(28, 14, 36, 0.95); border: 1px solid rgba(255, 20, 147, 0.3); border-radius: 8px; padding: 20px; margin: 20px 0; } .name-legend h3 { color: #FF1493; font-size: 18px; margin: 0 0 15px 0; } .legend-item { display: flex; align-items: baseline; gap: 10px; } .legend-key { color: #00FFEE; font-weight: 600; min-width: 80px; } .model-description { background: rgba(28, 14, 36, 0.95); border: 1px solid rgba(255, 20, 147, 0.3); border-radius: 8px; padding: 20px; } .model-description p { margin: 0 0 15px 0; line-height: 1.6; } .model-description p:last-child { margin-bottom: 0; } .info-card { background: rgba(28, 14, 36, 0.95); border: 1px solid rgba(255, 20, 147, 0.3); border-radius: 8px; overflow: hidden; } .info-header { background: rgba(255, 20, 147, 0.1); padding: 20px; border-bottom: 1px solid rgba(255, 20, 147, 0.3); } .info-header h3 { color: #FF1493; margin: 0 0 10px 0; font-size: 20px; text-shadow: 0 0 5px rgba(255, 20, 147, 0.3); } .model-tags { display: flex; gap: 8px; flex-wrap: wrap; } .model-tag { background: rgba(0, 255, 238, 0.1); color: #00FFEE; padding: 4px 8px; border-radius: 4px; font-size: 12px; border: 1px solid rgba(0, 255, 238, 0.2); } .model-composition { padding: 20px; border-bottom: 1px solid rgba(255, 20, 147, 0.3); } .model-composition h4 { color: #FF1493; margin: 0 0 15px 0; font-size: 16px; } .composition-list { list-style: none; padding: 0; margin: 0; display: grid; gap: 10px; } .composition-list li { color: #FFE1FF; display: flex; align-items: baseline; gap: 8px; } .model-component { color: #00FFEE; font-weight: 500; min-width: 120px; } .template-card { background: rgba(28, 14, 36, 0.95); border: 1px solid rgba(255, 20, 147, 0.3); border-radius: 8px; padding: 15px; } .template-item { display: flex; align-items: center; gap: 12px; } .template-icon { width: 24px; height: 24px; opacity: 0.8; } .template-content { display: flex; align-items: baseline; gap: 8px; } .template-link { color: #00FFEE; text-decoration: none; font-weight: 500; display: flex; align-items: center; gap: 5px; } .template-author { color: rgba(255, 225, 255, 0.7); font-size: 14px; } .quantized-section { background: rgba(28, 14, 36, 0.95); border: 1px solid rgba(255, 20, 147, 0.3); border-radius: 8px; padding: 20px; } .quantized-section h3 { color: #FF1493; font-size: 18px; margin: 0 0 15px 0; } .quantized-item { display: flex; align-items: baseline; gap: 10px; } .quantized-item .author { color: rgba(255, 225, 255, 0.7); min-width: 100px; } .multi-links { display: flex; align-items: center; gap: 8px; } .config-container { background: rgba(28, 14, 36, 0.95); border: 1px solid rgba(255, 20, 147, 0.3); border-radius: 8px; overflow: hidden; } .config-header { background: rgba(255, 20, 147, 0.1); padding: 15px 20px; border-bottom: 1px solid rgba(255, 20, 147, 0.3); } .config-item { display: flex; flex-direction: column; gap: 5px; margin-bottom: 15px; } .config-label { color: #00FFEE; font-size: 14px; font-weight: 500; } .config-value { color: #FFE1FF; font-family: 'Courier New', monospace; } .model-list { list-style: none; padding: 0; margin: 10px 0 0 0; } .model-list li { color: #FFE1FF; font-family: 'Courier New', monospace; padding: 5px 0; padding-left: 20px; position: relative; } .model-list li::before { content: '-'; position: absolute; left: 0; color: #00FFEE; } .link-arrow { display: inline-block; transition: transform 0.3s ease; } a:hover .link-arrow { transform: translateX(3px); } .benchmark-notification { background: rgba(255, 20, 147, 0.15); border: 1px solid rgba(255, 20, 147, 0.3); border-radius: 8px; margin-bottom: 20px; padding: 12px; animation: glowPulse 2s infinite; } .notification-content { display: flex; align-items: center; justify-content: center; gap: 10px; text-align: center; } .notification-text { color: #FFE1FF; font-size: 16px; font-weight: 500; display: flex; flex-direction: column; align-items: center; gap: 5px; } .benchmark-link { color: #00FFEE; text-decoration: none; font-size: 14px; padding: 4px 8px; border-radius: 4px; transition: all 0.3s ease; border: 1px solid rgba(0, 255, 238, 0.3); } .benchmark-link:hover { background: rgba(0, 255, 238, 0.1); border-color: rgba(0, 255, 238, 0.5); color: #00FFEE; text-shadow: 0 0 5px rgba(0, 255, 238, 0.5); } @keyframes glowPulse { 0% { box-shadow: 0 0 5px rgba(255, 20, 147, 0.3); } 50% { box-shadow: 0 0 15px rgba(255, 20, 147, 0.5); } 100% { box-shadow: 0 0 5px rgba(255, 20, 147, 0.3); } } .review-card { background: rgba(28, 14, 36, 0.95); border: 1px solid rgba(255, 20, 147, 0.3); border-radius: 8px; padding: 15px; margin-bottom: 15px; } suayptalha/Maestro-10B arcee-ai/Virtuoso-Lite DeepSeek-V3 10b Parameters Maestro-10B is a 10 billion parameter model fine-tuned from Virtuoso-Lite, a next-generation language model developed by arcee-ai. Virtuoso-Lite itself is based on the Llama-3 architecture, distilled from Deepseek-v3 using approximately 1.1 billion tokens/logits. This distillation process allows Virtuoso-Lite to achieve robust performance with a smaller parameter count, excelling in reasoning, code generation, and mathematical problem-solving. Maestro-10B inherits these strengths from its base model, Virtuoso-Lite, and further enhances them through fine-tuning on the OpenOrca dataset. This combination of a distilled base model and targeted fine-tuning makes Maestro-10B a powerful and efficient language model.

suayptalha/SilkGemma-2-9B-Q80-GGUF This model was converted to GGUF format from `suayptalha/SilkGemma-2-9B` using llama.cpp via the ggml.ai's GGUF-my-repo space. Refer to the original model card for more details on the model. Use with llama.cpp Install llama.cpp through brew (works on Mac and Linux) Note: You can also use this checkpoint directly through the usage steps listed in the Llama.cpp repo as well. Step 2: Move into the llama.cpp folder and build it with `LLAMACURL=1` flag along with other hardware-specific flags (for ex: LLAMACUDA=1 for Nvidia GPUs on Linux).

- Developed by: suayptalha - License: apache-2.0 - Finetuned from model : unsloth/gemma-2-9b-it-bnb-4bit This gemma2 model was trained 2x faster with Unsloth and Huggingface's TRL library.

This is a merge of pre-trained language models created using mergekit. This model was merged using the SLERP merge method. The following models were included in the merge: Xiaojian9992024/Qwen2.5-Dyanka-7B-Preview gz987/qwen2.5-7b-cabs-v0.3 The following YAML configuration was used to produce this model:

This is a merge of pre-trained language models created using mergekit. This model was merged using the SLERP merge method. The following models were included in the merge: Sakalti/ultiima-32B rombodawg/Rombos-LLM-V2.5-Qwen-32b The following YAML configuration was used to produce this model:

This project performs full fine-tuning on the Qwen3-0.6B language model to enhance its clinical diagnosis interpretation and reasoning capabilities. The model was optimized using the bfloat16 (bf16) data type. Dataset: Containing paired clinical patient histories and step-by-step diagnostic conclusions. Base model: Qwen3-0.6B, loaded with the `unsloth` library in bf16 precision. Full fine-tuning (`fullfinetuning=True`) applied to all layers to adapt the model for medical diagnostic tasks. Utilized the Hugging Face TRL library with the Supervised Fine-Tuning approach. The model was trained to generate both intermediate reasoning steps and final diagnostic statements. Training hyperparameters: Significantly improved the model’s ability to interpret clinical information and propose accurate, structured diagnoses. Performance was measured on a held-out validation set with the following metric: Diagnostic Similarity: 71.68% similarity compared to DeepSeek V3-0324 baseline. This project is licensed under the Apache License 2.0. See the LICENSE file for details.

This project performs full fine-tuning on the Qwen3-0.6B language model to enhance its clinical treatment planning and reasoning capabilities. The model was optimized using the bfloat16 (bf16) data type. Dataset: Containing paired clinical diagnosis descriptions and corresponding step-by-step treatment plans. Base model: Qwen3-0.6B, loaded with the `unsloth` library in bf16 precision. Full fine-tuning (`fullfinetuning=True`) applied to all layers to adapt the model for medical treatment tasks. Utilized the Hugging Face TRL library with the Supervised Fine-Tuning approach. The model was trained to generate both intermediate reasoning steps and final treatment recommendations. Training hyperparameters: Significantly improved the model’s ability to interpret clinical diagnoses and propose structured treatment plans. Performance was measured on a held-out validation set with the following metrics: Plan Fidelity: 59.69% similarity with DeepSeek V3-0324. Reasoning Coherence: Rated high by a panel of medical experts. License This project is licensed under the Apache License 2.0. See the LICENSE file for details.