LocalLLaMA/local-model-explorer-data
Viewer โข Updated โข 4.36k โข 6.59k โข 3
AI & ML interests
Local LLMs
Recent Activity
codelionย
updated a
dataset about 8 hours ago
gianlucamazzaย
posted an update about 22 hours ago
Post
61
A keyless, offline playground for mklang, my language and interpreter for LLM control flow. Parse, check and lint, format, and run the bundled examples and conformance cases. No model runs and no API key is needed; judge picks come from fixture files.
gianlucamazza/mklang-playground
gianlucamazza/mklang-playground
gianlucamazzaย
posted an update 2 days ago
Post
78
My public work on the Hub: the emotional-memory demo and benchmark dataset, the mklang playground, the affective-fly demo, and an int4 ONNX Runtime GenAI copy of SmolLM2-360M-Instruct. Each card states what it is for and where it stops working. The SmolLM2 copy is experimental, with no quality evaluation.
https://huggingface.co/collections/gianlucamazza/agent-memory-affect-mklang-and-local-inference
https://huggingface.co/collections/gianlucamazza/agent-memory-affect-mklang-and-local-inference
MaziyarPanahiย
posted an update 3 days ago
Post
189
Used nvidia/NV-Reason-CT by Nvidia and Hugging Face Jobs to make this cool demo! Let me know if you think we should add more features to it: ๐ฅ
https://www.linkedin.com/feed/update/urn:li:activity:7515789448077398016
https://www.linkedin.com/feed/update/urn:li:activity:7515789448077398016
Submission: FINAL-Bench/Darwin-27B-ZTC (general, zero-shot), accuracy 0.743
๐ฅ 1
3
#17 opened 9 days ago
by 
SeaWolf-AI
codelionย
updated a
dataset 3 days ago
Post
85
zixi-51M: Not a stock voice. A voice. ๐พ
NoesisLab/zixi-51m
Everyone is shipping 500B-parameter models. I shipped 51 million. With one voice. On purpose. ๐ค
What: open-weights Chinese TTS. VITS2 lineage. 51.43M params. Runs on your laptop CPU โ no cloud, no API key, no GPU bill, no meter running.๐ธ
The voice: not a stock voice, not a 5,000-speaker average. One real human's voice โ mine. Timbre, rhythm, breath included. The sample on themodel card is one take, default parameters, recorded by absolutely no one. ๐ง
What's NOT inside:
โข โ BERT (zero. the text encoder vibes on phones + tones alone)
โข โ language routing (Chinese-exclusive: 77 symbols, 6 tones, one job)
โข โ my training data (the corpus is my own writing and stays mine โ open weights, closed corpus, that's the deal ๐ค)
The trick: embedding transplant. Phone, tone and speaker rows inherited from the MeloTTS-Chinese base by symbol name โ no random init. Themodel woke up already knowing Chinese. ๐ง
Built solo, corpus to packaging, by a writer who is also a violinist. Prosody is my day job. ๐ป
Use it, fork it, fine-tune your own voice. Just don't put words in a real person's mouth. ๐
๐ Weights: NoesisLab/zixi-51m
๐ป Code: https://github.com/lizixi-0x2F/zixi-51M
NoesisLab/zixi-51m
Everyone is shipping 500B-parameter models. I shipped 51 million. With one voice. On purpose. ๐ค
What: open-weights Chinese TTS. VITS2 lineage. 51.43M params. Runs on your laptop CPU โ no cloud, no API key, no GPU bill, no meter running.๐ธ
The voice: not a stock voice, not a 5,000-speaker average. One real human's voice โ mine. Timbre, rhythm, breath included. The sample on themodel card is one take, default parameters, recorded by absolutely no one. ๐ง
What's NOT inside:
โข โ BERT (zero. the text encoder vibes on phones + tones alone)
โข โ language routing (Chinese-exclusive: 77 symbols, 6 tones, one job)
โข โ my training data (the corpus is my own writing and stays mine โ open weights, closed corpus, that's the deal ๐ค)
The trick: embedding transplant. Phone, tone and speaker rows inherited from the MeloTTS-Chinese base by symbol name โ no random init. Themodel woke up already knowing Chinese. ๐ง
Built solo, corpus to packaging, by a writer who is also a violinist. Prosody is my day job. ๐ป
Use it, fork it, fine-tune your own voice. Just don't put words in a real person's mouth. ๐
๐ Weights: NoesisLab/zixi-51m
๐ป Code: https://github.com/lizixi-0x2F/zixi-51M
Amos 1 medium and large: general zero-shot scores, plus the large fitted on train
๐ฅ 1
1
#31 opened 4 days ago
by 
mukaddamzaid
Submission: rony/ministral3-3b-reasoning-typed-decisions (Specialist) (Ministral 3 3B Reasoning, fine-tuned on train)
1
#30 opened 4 days ago
by 
rony
Add Amos 1 medium and large (general, zero-shot) and the large's fitted row
1
#32 opened 3 days ago
by mukaddamzaid
Add jev-vision-27b (general, zero-shot): acc 0.714, KL 0.318
๐ฅ๐ 1
2
#28 opened 4 days ago
by 
e12ex2
Add ProKope-421M-8K (Alan Canto) to Table 2 (71.85% acc, 8K context)
1
#24 opened 5 days ago
by 
AlanCantoFTW
Submission: rony/gemma3-270m-typed-decisions (Specialist)
1
#27 opened 5 days ago
by rony
Submission: rony/gemma3-270m-typed-decisions (Gemma 3 270M, fine-tuned on train)[General]
1
#26 opened 5 days ago
by rony
Submission: ProKope-421M-8K (Alan Canto) โ 0.7185 acc / 0.385 Brier / 8K context (specialist)
๐ฅ 1
1
#23 opened 5 days ago
by AlanCantoFTW
Qwen3.5-4B LoRA fine-tune: results submitted, model missing from the Hub leaderboard
๐ฅ 1
1
#22 opened 6 days ago
by 
wozniakos10
gianlucamazzaย
posted an update 5 days ago
Post
4,541
Upstream models my open-source projects run or my articles discuss. Each note says which it is. thenlper/gte-small is real use: RAGFS embeds file chunks with it locally. The two retrieval models from my RAG article, a cross-encoder re-ranker and a bi-encoder, are examples, labelled as such, not claims of production use.
gianlucamazza/models-used-in-my-projects-6aca0b3201f38f4c264ac9bf
gianlucamazza/models-used-in-my-projects-6aca0b3201f38f4c264ac9bf
codelionย
updated a
Space 6 days ago