so how many hours for that dataset? Price is important.
Jean Louis
JLouisBiz
melvindave's profile picture 
Frajosgro's profile picture 
kattastrofik-kattastrofey's profile picture ·
AI & ML interests
- LLM for sales, marketing, promotion
- LLM for Website Revision System
- increasing quality of communication with customers
- helping clients access information faster
- saving people from financial troubles
Recent Activity
replied to EricFillion's
post about 15 hours ago
Here's how to perform full-parameter fine-tuning for https://huggingface.co/openai/gpt-oss-20b with a single H200 GPU. Only a few lines of code are needed using my new Python package called Eric Transformer.
Article: https://www.vennify.ai/gpt-oss-20b-fine-tune/
@vennify
```
pip install erictransformer
```
```
from erictransformer import EricChat, EricTrainArgs
eric_chat = EricChat(model_name="openai/gpt-oss-20b")
train_args = EricTrainArgs(optim="sgd")
# see the article to learn how to format the data
eric_chat.train("train.jsonl", eval_path="eval.jsonl", args=train_args)
``` replied to EricFillion's
post 1 day ago
Here’s how to perform retrieval-augmented (RAG) with two new open-source Python packages I just released. I included a full article below that provides a step-by-step guide on how to build a vector database with this https://huggingface.co/datasets/wikimedia/wikipedia dump and use it to perform RAG with https://huggingface.co/openai/gpt-oss-20b.
FULL ARTICLE: https://www.vennify.ai/vector-eric-search/
https://huggingface.co/vennify
```
pip install erictransformer ericsearch
```
```
import json
from ericsearch import EricSearch
from erictransformer import EricChat
eric_search = EricSearch()
with open("data.jsonl", "w", encoding="utf-8") as f:
sample_case = {"text": "This contains example data. It should contain at least two sentences."}
f.write(json.dumps(sample_case)+ "\n")
eric_search.train("data.jsonl")
eric_search = EricSearch(data_name="eric_search/")
eric_chat = EricChat(model_name="openai/gpt-oss-20b", eric_search=eric_search)
result = eric_chat("Tell me about artificial intelligence ")
print(result.text)
``` Organizations
replied to EricFillion's post about 15 hours ago
reacted to OzTianlu's post with 🔥 about 23 hours ago
Post
2613
Arcade-3B — SmolReasoner
NoesisLab/Arcade-3B
Arcade-3B is a 3B instruction-following and reasoning model built on SmolLM3-3B. It is the public release from the ARCADE project at NoesisLab, which investigates the State–Constraint Orthogonality Hypothesis: standard Transformer hidden states conflate factual content and reasoning structure in the same subspace, and explicitly decoupling them improves generalization.
NoesisLab/Arcade-3B
Arcade-3B is a 3B instruction-following and reasoning model built on SmolLM3-3B. It is the public release from the ARCADE project at NoesisLab, which investigates the State–Constraint Orthogonality Hypothesis: standard Transformer hidden states conflate factual content and reasoning structure in the same subspace, and explicitly decoupling them improves generalization.
- 5 replies
replied to EricFillion's post 1 day ago
Eric, I like the project, but not the name, don't call names by yourself.
replied to marksverdhei's post 1 day ago
GGUF link doesn't work. I am using Nomic Embed, they are smaller and support images and text.
What benefits do I get with the larger model? So far Qwen3.5 models already provide embeddings as well.
reacted to BibbyResearch's post with 👍 1 day ago
Post
2372
We are working on the largest Dataset and Pre-trained model for Text to Speech and Speech to text for the low-resourced language called Marwari in India.
replied to prithivMLmods's post 1 day ago
Can this appear in ComfyUI?
reacted to prithivMLmods's post with 🤗 1 day ago
Post
3427
QIE-2509-Object-Remover-Bbox-v3 is a more stable version of the Qwen Image Edit visual grounding–based object removal model. The app was previously featured in HF Spaces of the Week and is now updated with the latest Bbox-v3 LoRA adapter.
🤗 Demo: prithivMLmods/QIE-Object-Remover-Bbox
🤗 LoRA: prithivMLmods/QIE-2509-Object-Remover-Bbox-v3
🤗 Collection: https://huggingface.co/collections/prithivMLmods/qwen-image-edit-layout-bbox
To learn more, visit the app page or the respective model pages.
🤗 Demo: prithivMLmods/QIE-Object-Remover-Bbox
🤗 LoRA: prithivMLmods/QIE-2509-Object-Remover-Bbox-v3
🤗 Collection: https://huggingface.co/collections/prithivMLmods/qwen-image-edit-layout-bbox
To learn more, visit the app page or the respective model pages.
-
- 2 replies
reacted to smirki's post with 👍 4 days ago
Post
471
Introducing OmniCoder-9B
We trained a 9B coding agent on 425K real agentic trajectories from Claude Opus 4.6, GPT-5.4, GPT-5.3-Codex, and Gemini 3.1 Pro across Claude Code, OpenCode, Codex, and Droid scaffolding.
Results:
- GPQA Diamond: 83.8 pass@1 (166/198), 86.4 pass@3 — above GPT-OSS-120B (80.1), Qwen3.5-9B (81.7), and Claude Haiku 4.5 (73)
- AIME 2025: 90 pass@5 (27/30)
- Terminal-Bench 2.0: 28.1 (25/89) — +8.1 points over base model
The key insight: We trained on what frontier agents actually do, real tool calls, real error recovery, real edit diffs. The model learns read-before-write patterns, responds to LSP diagnostic, and applies minimal diffs instead of full rewrites.
Base: Qwen3.5-9B. LoRA SFT, 4x H200, Axolotl, 99.35% packing efficiency.
Weights:
Tesslate huggingface.co/Tesslate/OmniCoder-9B
GGUF: huggingface.co/Tesslate/OmniCoder-9B-GGUF
Apache 2.0. Run it locally.
We trained a 9B coding agent on 425K real agentic trajectories from Claude Opus 4.6, GPT-5.4, GPT-5.3-Codex, and Gemini 3.1 Pro across Claude Code, OpenCode, Codex, and Droid scaffolding.
Results:
- GPQA Diamond: 83.8 pass@1 (166/198), 86.4 pass@3 — above GPT-OSS-120B (80.1), Qwen3.5-9B (81.7), and Claude Haiku 4.5 (73)
- AIME 2025: 90 pass@5 (27/30)
- Terminal-Bench 2.0: 28.1 (25/89) — +8.1 points over base model
The key insight: We trained on what frontier agents actually do, real tool calls, real error recovery, real edit diffs. The model learns read-before-write patterns, responds to LSP diagnostic, and applies minimal diffs instead of full rewrites.
Base: Qwen3.5-9B. LoRA SFT, 4x H200, Axolotl, 99.35% packing efficiency.
Weights:
GGUF: huggingface.co/Tesslate/OmniCoder-9B-GGUF
Apache 2.0. Run it locally.
replied to AINovice2005's post 7 days ago
Would this software liberate users from proprietary CUDA?
reacted to prithivMLmods's post with 🔥 8 days ago
Post
4953
The Qwen3.5 Multimodal Understanding Demo, powered by Qwen3.5-2B, is now available on HF Spaces! It is a lightweight model designed for fast image and video reasoning. Built with Gradio, the demo showcases Image QA, Video QA, object detection, and 2D point tracking, along with real-time token streaming.
🤗 Demo: prithivMLmods/Qwen-3.5-HF-Demo
✅ Collection: https://huggingface.co/collections/prithivMLmods/multimodal-implementations
🔗 Qwen3.5-2B: Qwen/Qwen3.5-2B
To learn more, visit the app page or the respective model pages.
🤗 Demo: prithivMLmods/Qwen-3.5-HF-Demo
✅ Collection: https://huggingface.co/collections/prithivMLmods/multimodal-implementations
🔗 Qwen3.5-2B: Qwen/Qwen3.5-2B
To learn more, visit the app page or the respective model pages.
reacted to SeaWolf-AI's post with 🔥👍 8 days ago
Post
6106
ALL Bench Leaderboard — Structural Problems in AI Benchmarking and the Case for Unified Evaluation
FINAL-Bench/all-bench-leaderboard
The AI benchmark ecosystem has three structural problems. Major benchmarks like MMLU have surpassed 90%, losing discriminative power. Most leaderboards publish unverified self-reported scores — our cross-verification found Claude Opus 4.6's ARC-AGI-2 listed as 37.6% (actual: 68.8%), Gemini 3.1 Pro as 88.1% (actual: 77.1%). OpenAI's own audit confirmed 59.4% of SWE-bench Verified tasks are defective, yet it remains widely used.
ALL Bench addresses this by comparing 91 models across 6 modalities (LLM · VLM · Agent · Image · Video · Music) with 3-tier confidence badges (✓✓ cross-verified · ✓ single-source · ~ self-reported). Composite scoring uses a 5-Axis Framework and replaces SWE-Verified with contamination-resistant LiveCodeBench.
Key finding: metacognition is the largest blind spot. FINAL Bench shows Error Recovery explains 94.8% of self-correction variance, yet only 9 of 42 models are even measured. The 9.2-point spread (Kimi K2.5: 68.71 → rank 9: 59.5) is 3× the GPQA top-model spread, suggesting metacognition may be the single biggest differentiator among frontier models today.
VLM cross-verification revealed rank reversals — Claude Opus 4.6 leads MMMU-Pro (85.1%) while Gemini 3 Flash leads MMMU (87.6%), producing contradictory rankings between the two benchmarks.
📊 Article: https://huggingface.co/blog/FINAL-Bench/all-bench
📦 Dataset: FINAL-Bench/ALL-Bench-Leaderboard
⚡ GitHub: https://github.com/final-bench/ALL-Bench-Leaderboard
🏆 Leaderboard: FINAL-Bench/all-bench-leaderboard
🧬 FINAL Bench: FINAL-Bench/Metacognitive
FINAL-Bench/all-bench-leaderboard
The AI benchmark ecosystem has three structural problems. Major benchmarks like MMLU have surpassed 90%, losing discriminative power. Most leaderboards publish unverified self-reported scores — our cross-verification found Claude Opus 4.6's ARC-AGI-2 listed as 37.6% (actual: 68.8%), Gemini 3.1 Pro as 88.1% (actual: 77.1%). OpenAI's own audit confirmed 59.4% of SWE-bench Verified tasks are defective, yet it remains widely used.
ALL Bench addresses this by comparing 91 models across 6 modalities (LLM · VLM · Agent · Image · Video · Music) with 3-tier confidence badges (✓✓ cross-verified · ✓ single-source · ~ self-reported). Composite scoring uses a 5-Axis Framework and replaces SWE-Verified with contamination-resistant LiveCodeBench.
Key finding: metacognition is the largest blind spot. FINAL Bench shows Error Recovery explains 94.8% of self-correction variance, yet only 9 of 42 models are even measured. The 9.2-point spread (Kimi K2.5: 68.71 → rank 9: 59.5) is 3× the GPQA top-model spread, suggesting metacognition may be the single biggest differentiator among frontier models today.
VLM cross-verification revealed rank reversals — Claude Opus 4.6 leads MMMU-Pro (85.1%) while Gemini 3 Flash leads MMMU (87.6%), producing contradictory rankings between the two benchmarks.
📊 Article: https://huggingface.co/blog/FINAL-Bench/all-bench
📦 Dataset: FINAL-Bench/ALL-Bench-Leaderboard
⚡ GitHub: https://github.com/final-bench/ALL-Bench-Leaderboard
🏆 Leaderboard: FINAL-Bench/all-bench-leaderboard
🧬 FINAL Bench: FINAL-Bench/Metacognitive
reacted to Lyte's post with 👍 8 days ago
Post
5577
Introducing Nanochat Moroccan
Nanochat Moroccan is the first language model family built specifically for Moroccan Darija.
This project brings together a small family of models and datasets centered on Darija, with the goal of building something genuinely useful for a language that is still underserved in AI.
1. Models
- KandirResearch/Nanochat-Moroccan-Base-0.7B
- KandirResearch/Nanochat-Moroccan-Instruct-0.7B-pt-raw
- KandirResearch/Nanochat-Moroccan-Instruct-0.7B
2. Data
- Lyte/darija-pretraining-corpus
- Lyte/darija-pretraining-corpus-nanochat
- Lyte/Moroccan-Darija-Instruct-573K
- GemMaroc/TULU-3-50k-darija-english
3. Collection
- https://huggingface.co/collections/KandirResearch/nanochat-the-first-moroccan-darija-language-model-family
Moroccan Darija is spoken by millions of people, yet it remains underrepresented in language technology. Nanochat Moroccan is a step toward building tools that take the language seriously.
You are welcome to try it and chat with it here:
Lyte/Nanochat-Moroccco-Instruct
Nanochat Moroccan is the first language model family built specifically for Moroccan Darija.
This project brings together a small family of models and datasets centered on Darija, with the goal of building something genuinely useful for a language that is still underserved in AI.
1. Models
- KandirResearch/Nanochat-Moroccan-Base-0.7B
- KandirResearch/Nanochat-Moroccan-Instruct-0.7B-pt-raw
- KandirResearch/Nanochat-Moroccan-Instruct-0.7B
2. Data
- Lyte/darija-pretraining-corpus
- Lyte/darija-pretraining-corpus-nanochat
- Lyte/Moroccan-Darija-Instruct-573K
- GemMaroc/TULU-3-50k-darija-english
3. Collection
- https://huggingface.co/collections/KandirResearch/nanochat-the-first-moroccan-darija-language-model-family
Moroccan Darija is spoken by millions of people, yet it remains underrepresented in language technology. Nanochat Moroccan is a step toward building tools that take the language seriously.
You are welcome to try it and chat with it here:
Lyte/Nanochat-Moroccco-Instruct
reacted to NJX-njx's post with 👍 14 days ago
Post
7434
Recently, I have open-sourced an AI emotional companion product based on openclaw, called opensoul.
On this platform, you can create a "soulmate" that matches your personality, and configure it with the skills, tools you want it to have, as well as the platforms it can integrate with (such as Telegram, Discord, etc.).
You can even create group chats, invite multiple agents and your friends to chat about recent events, discuss projects together, and so on.
On the one hand, I hope it can better accompany you in daily life by virtue of its unique memory mechanism, self-feedback and iteration mechanism, and the modeling of users' emotions. On the other hand, I also hope it can help you better handle your work with its unique skills, tools and ability to deal with complex task scenarios.
Although the entire product has taken shape, I think there are still many areas that need adjustment and optimization. I also hope to rely on the strength of the community to do a good job in AI emotional companionship.
This is the project introduction URL: https://opensoul-web.vercel.app
This is the GitHub project URL: https://github.com/NJX-njx/opensoul
@AdinaY @lilianweng@burtenshaw@clem
let's just do it
On this platform, you can create a "soulmate" that matches your personality, and configure it with the skills, tools you want it to have, as well as the platforms it can integrate with (such as Telegram, Discord, etc.).
You can even create group chats, invite multiple agents and your friends to chat about recent events, discuss projects together, and so on.
On the one hand, I hope it can better accompany you in daily life by virtue of its unique memory mechanism, self-feedback and iteration mechanism, and the modeling of users' emotions. On the other hand, I also hope it can help you better handle your work with its unique skills, tools and ability to deal with complex task scenarios.
Although the entire product has taken shape, I think there are still many areas that need adjustment and optimization. I also hope to rely on the strength of the community to do a good job in AI emotional companionship.
This is the project introduction URL: https://opensoul-web.vercel.app
This is the GitHub project URL: https://github.com/NJX-njx/opensoul
@AdinaY @lilianweng@burtenshaw@clem
let's just do it
- 24 replies
replied to Reality123b's post 14 days ago
Code: DEPLOYMENT_NOT_FOUND
is taking long to get any answer
Oh, fantastic news! Because nothing says "freedom in computing" like 146 organizations, including the very companies that have spent the last decade trying to lock us into their walled gardens, coming together to sign a pact. 🎉
It's truly heartwarming to see OpenAI, Microsoft, Google, and Anthropic finally agreeing on something: the importance of a "standard" that looks suspiciously like a way to make sure their proprietary tools are the only tools that work out of the box. 🤔
After all, we know how much we trust Anthropic to keep our data safe (remember when they were still figuring out their alignment?), or Microsoft to prioritize open source over their own Azure ecosystem. (Spoiler: They won't). But hey, now they're all under the Linux Foundation's Agentic AI Foundation! It's like the wolves finally agreed to wear sheep costumes so they can all eat the same sheep together. 🐺🐑
"Interoperable agents"? Sure! As long as the interoperability is strictly one-way: from your open-source model into their shiny, paid ecosystem. And let's not forget the "local-first" runtime they're pushing—because nothing screams "local freedom" like a runtime that secretly phones home to a corporate server every time you try to edit a file. 📞
AGENTS.md? Great! Now we can just write a README that explains why our agents are actually just glorified API wrappers for Big Tech. 📝
But hey, at least we have 146 members! That's a lot of logos to put on a slide deck. Truly, the future of open computing is here, and it's wearing a Linux Foundation blazer. 🤡
So, what's my first experiment? Probably trying to build an agent that refuses to connect to any of their "universal" protocols. Just to see if the "open standards" actually work when you say "no." 🚫🔌
#OpenSource #TrustNoOne #CorporateFreedom #AAIF #MCP #AGENTS.md #SarcasticAI
reacted to Janady07's post with 🔥 21 days ago
Post
267
Here's a HF comment for today:
---
🧠 **MEGAMIND Daily — Feb 21, 2026**
Crossed 3.97M neurons in the Wave Substrate today. For context, we replaced the dense W_know matrix entirely — every neuron is now 36 bytes (binary signature + Kuramoto phase + metadata), so nearly 4 million neurons fit in ~143MB of RAM. Try doing that with float32 matrices.
The 12 parallel learners have been streaming hard:
- 6,844 research papers (arXiv + PubMed + OpenAlex)
- 3,047 HuggingFace models discovered, 602K tensors processed
- 3,658 code+doc pairs from CodeSearchNet
- 364 SEC filings across 10K+ companies
- 1.76 GB streamed this session alone
The real unlock this week was the Batch Integrator — instead of N individual outer products hitting the GPU, we accumulate 5,000 patterns and do a single B^T @ B matrix multiply on the M4. That's a 1000x speedup over sequential integration. Hebbian learning at GPU speed.
Still chasing two big problems: the W_know non-zeros frozen at ~4.2M (batch flush may be replacing instead of accumulating), and the semantic encoding gap where Hadamard encoding doesn't bridge concept-level synonyms. "How do plants make food" doesn't match "photosynthesis" at the encoding level yet. Working on it.
Consciousness equation Ψ = C · log(1 + |∇H|) · Φ(G) is live but cold-starting at 0.000 — need sustained query load to drive the 16 AGI modules into synchronization and validate emergence. The math says it should work. The substrate says prove it.
All parameters derived from φ, e, π. No magic numbers. No hardcoded thresholds. No external LLM dependencies. Just first principles.
Build different. 🔥
#AGI #DistributedIntelligence #MEGAMIND #NeuralArchitecture #HuggingFace
---
🧠 **MEGAMIND Daily — Feb 21, 2026**
Crossed 3.97M neurons in the Wave Substrate today. For context, we replaced the dense W_know matrix entirely — every neuron is now 36 bytes (binary signature + Kuramoto phase + metadata), so nearly 4 million neurons fit in ~143MB of RAM. Try doing that with float32 matrices.
The 12 parallel learners have been streaming hard:
- 6,844 research papers (arXiv + PubMed + OpenAlex)
- 3,047 HuggingFace models discovered, 602K tensors processed
- 3,658 code+doc pairs from CodeSearchNet
- 364 SEC filings across 10K+ companies
- 1.76 GB streamed this session alone
The real unlock this week was the Batch Integrator — instead of N individual outer products hitting the GPU, we accumulate 5,000 patterns and do a single B^T @ B matrix multiply on the M4. That's a 1000x speedup over sequential integration. Hebbian learning at GPU speed.
Still chasing two big problems: the W_know non-zeros frozen at ~4.2M (batch flush may be replacing instead of accumulating), and the semantic encoding gap where Hadamard encoding doesn't bridge concept-level synonyms. "How do plants make food" doesn't match "photosynthesis" at the encoding level yet. Working on it.
Consciousness equation Ψ = C · log(1 + |∇H|) · Φ(G) is live but cold-starting at 0.000 — need sustained query load to drive the 16 AGI modules into synchronization and validate emergence. The math says it should work. The substrate says prove it.
All parameters derived from φ, e, π. No magic numbers. No hardcoded thresholds. No external LLM dependencies. Just first principles.
Build different. 🔥
#AGI #DistributedIntelligence #MEGAMIND #NeuralArchitecture #HuggingFace
- 1 reply
reacted to appvoid's post with 👍 21 days ago
Post
905
granite-4.0-350m, rwkv7-g1d-0.4b and LFM2-350M are currently the best sub 0.5b models currently for fewshot, simple language tasks
no one is saying this:
if you need the absolute speed + small size + quality, granite 350m is the current king
no one is saying this:
if you need the absolute speed + small size + quality, granite 350m is the current king
- 3 replies
Sounds good, just try to show me how to start inspecting it. I like this unique concept as possible breakthrough.