Domain Adaptation of Llama3-70B-Instruct through Continual Pre-Training and Model Merging: A Comprehensive Evaluation
Paper
• 2406.14971 • Published
AI & ML interests
None defined yet.
Recent Activity
View all activity
anneketh-vij
authored 2 papers about 4 hours ago
Alyosha11
authored a paper 3 days ago
pranavkr
authored a paper 3 days ago
arthurbresnu
authored a paper 3 days ago
MaziyarPanahi
authored a paper 3 days ago
Jackmin108
authored a paper 3 days ago
chargoddard
authored a paper 3 days ago
danielhanchen
posted an
update 4 days ago
Post
2407
We collabed with HF on showing how you can use HF Jobs and Unsloth! https://huggingface.co/blog/unsloth-jobs
danielhanchen
posted an
update 7 days ago
Post
5557
You can now run Qwen3.5 locally! 💜
Qwen3.5-397B-A17B is an open MoE vision reasoning LLM for agentic coding & chat. It performs on par with Gemini 3 Pro, Claude Opus 4.5 & GPT-5.2.
GGUF: unsloth/Qwen3.5-397B-A17B-GGUF
Run Dynamic 3-bit on a 192GB Mac for 20 tokens/s.
Guide: https://unsloth.ai/docs/models/qwen3.5
Qwen3.5-397B-A17B is an open MoE vision reasoning LLM for agentic coding & chat. It performs on par with Gemini 3 Pro, Claude Opus 4.5 & GPT-5.2.
GGUF: unsloth/Qwen3.5-397B-A17B-GGUF
Run Dynamic 3-bit on a 192GB Mac for 20 tokens/s.
Guide: https://unsloth.ai/docs/models/qwen3.5
- 9 replies
danielhanchen
posted an
update 8 days ago
Post
8376
You can now run MiniMax-2.5 locally! 🚀
At 230B parameters, MiniMax-2.5 is the strongest LLM under 700B params, delivering SOTA agentic coding & chat.
Run Dynamic 3/4-bit on a 128GB Mac for 20 tokens/s.
Guide: https://unsloth.ai/docs/models/minimax-2.5
GGUF: unsloth/MiniMax-M2.5-GGUF
At 230B parameters, MiniMax-2.5 is the strongest LLM under 700B params, delivering SOTA agentic coding & chat.
Run Dynamic 3/4-bit on a 128GB Mac for 20 tokens/s.
Guide: https://unsloth.ai/docs/models/minimax-2.5
GGUF: unsloth/MiniMax-M2.5-GGUF
- 1 reply
MaziyarPanahi
posted an
update 13 days ago
Post
1795
Announcing: OpenMed Multilingual PII Detection Models
Today I am releasing 105 open-source models for Personally Identifiable Information (PII) detection in French, German, and Italian.
All Apache 2.0 licensed. Free for commercial use. No restrictions.
Performance:
- French: 97.97% F1 (top model)
- German: 97.61% F1 (top model)
- Italian: 97.28% F1 (top model)
All top-10 models per language exceed 96% F1
Coverage:
55+ PII entity types per language
Native ID formats: NSS (French), Sozialversicherungsnummer (German), Codice Fiscale (Italian)
Language-specific address, phone, and name patterns
Training Data:
French: 49,580 samples
German: 42,250 samples
Italian: 40,944 samples
Why Multilingual?
European healthcare operates in European languages. Clinical notes, patient records, and medical documents are generated in French, German, Italian, and other languages.
Effective de-identification requires:
- Native language understanding — not translation
- Local ID format recognition — each country has unique patterns
- Cultural context awareness — names, addresses, and formats vary
- These models deliver production-ready accuracy without requiring data to leave your infrastructure or language.
HIPAA & GDPR Compliance
Built for US and European privacy regulations:
- On-premise deployment: Process data locally with zero external dependencies
- Data sovereignty: No API calls, no cloud services, no cross-border transfers
- Air-gapped capable: Deploy in fully isolated environments if required
- Regulatory-grade accuracy: Supporting Expert Determination standards
- HIPAA and GDPR compliance across languages, without compliance gaps.
Use Cases
- Hospital EHR systems: Automated patient record de-identification
- Clinical research: Multilingual dataset preparation for studies
- Insurance companies: Claims processing across
https://huggingface.co/collections/OpenMed/multilingual-pii-and-de-identification
Today I am releasing 105 open-source models for Personally Identifiable Information (PII) detection in French, German, and Italian.
All Apache 2.0 licensed. Free for commercial use. No restrictions.
Performance:
- French: 97.97% F1 (top model)
- German: 97.61% F1 (top model)
- Italian: 97.28% F1 (top model)
All top-10 models per language exceed 96% F1
Coverage:
55+ PII entity types per language
Native ID formats: NSS (French), Sozialversicherungsnummer (German), Codice Fiscale (Italian)
Language-specific address, phone, and name patterns
Training Data:
French: 49,580 samples
German: 42,250 samples
Italian: 40,944 samples
Why Multilingual?
European healthcare operates in European languages. Clinical notes, patient records, and medical documents are generated in French, German, Italian, and other languages.
Effective de-identification requires:
- Native language understanding — not translation
- Local ID format recognition — each country has unique patterns
- Cultural context awareness — names, addresses, and formats vary
- These models deliver production-ready accuracy without requiring data to leave your infrastructure or language.
HIPAA & GDPR Compliance
Built for US and European privacy regulations:
- On-premise deployment: Process data locally with zero external dependencies
- Data sovereignty: No API calls, no cloud services, no cross-border transfers
- Air-gapped capable: Deploy in fully isolated environments if required
- Regulatory-grade accuracy: Supporting Expert Determination standards
- HIPAA and GDPR compliance across languages, without compliance gaps.
Use Cases
- Hospital EHR systems: Automated patient record de-identification
- Clinical research: Multilingual dataset preparation for studies
- Insurance companies: Claims processing across
https://huggingface.co/collections/OpenMed/multilingual-pii-and-de-identification
-
- 1 reply
danielhanchen
posted an
update 13 days ago
Post
5129
We collaborated with Hugging Face to enable you to train MoE models 12× faster with 35% less VRAM via our new Triton kernels (no accuracy loss). 🤗
Train gpt-oss locally on 12.8GB VRAM with our free notebooks: https://unsloth.ai/docs/new/faster-moe
Train gpt-oss locally on 12.8GB VRAM with our free notebooks: https://unsloth.ai/docs/new/faster-moe
- 1 reply
MaziyarPanahi
posted an
update 16 days ago
Post
1203
From Golden Gate Bridge to Broken JSON: Why Anthropic's SAE Steering Fails for Structured Output
I ran 6 experiments trying to use Anthropic's SAE steering for JSON generation.
- Base model: 86.8% valid JSON
- Steering only: 24.4%
- Fine-tuned: 96.6%
- FSM constrained: 100%
Steering is for semantics, not syntax.
https://huggingface.co/blog/MaziyarPanahi/sae-steering-json
I ran 6 experiments trying to use Anthropic's SAE steering for JSON generation.
- Base model: 86.8% valid JSON
- Steering only: 24.4%
- Fine-tuned: 96.6%
- FSM constrained: 100%
Steering is for semantics, not syntax.
https://huggingface.co/blog/MaziyarPanahi/sae-steering-json
MaziyarPanahi
posted an
update 17 days ago
Post
3939
🚨 Day 8/8: OpenMed Medical Reasoning Dataset Release - THE GRAND FINALE
Today I complete my 8-day release series with Medical-Reasoning-SFT-Mega.
The largest open medical reasoning dataset, combining 7 state-of-the-art AI models with fair distribution deduplication.
THE 7 SOURCE MODELS (Original Sample Counts):
1. Trinity-Mini: 810,284 samples
2. Qwen3-Next-80B: 604,249 samples
3. GPT-OSS-120B: 506,150 samples
4. Nemotron-Nano-30B: 444,544 samples
5. GLM-4.5-Air: 225,179 samples
6. MiniMax-M2.1: 204,773 samples
7. Baichuan-M3-235B: 124,520 samples
TOTAL BEFORE DEDUPLICATION: 2,919,699 samples
TOKEN COUNTS:
- Content tokens: 2.22 Billion
- Reasoning tokens: 1.56 Billion
- Total tokens: 3.78 Billion
- Samples with chain-of-thought: 100%
Quick Start:
All datasets Apache 2.0 licensed. Free for research and commercial use.
Thank you for following OpenMed's release series. I can't wait to see what you build. 🔥
OpenMed/Medical-Reasoning-SFT-Mega
OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B-V2
OpenMed/Medical-Reasoning-SFT-Trinity-Mini
OpenMed/Medical-Reasoning-SFT-GLM_4.5_Air
OpenMed/Medical-Reasoning-SFT-MiniMax-M2.1
OpenMed/Medical-Reasoning-SFT-Qwen3-Next-80B
OpenMed/Medical-Reasoning-SFT-Nemotron-Nano-30B
https://huggingface.co/datasets/OpenMed/Medical-Reasonin
https://huggingface.co/collections/OpenMed/medical-datasets
Today I complete my 8-day release series with Medical-Reasoning-SFT-Mega.
The largest open medical reasoning dataset, combining 7 state-of-the-art AI models with fair distribution deduplication.
THE 7 SOURCE MODELS (Original Sample Counts):
1. Trinity-Mini: 810,284 samples
2. Qwen3-Next-80B: 604,249 samples
3. GPT-OSS-120B: 506,150 samples
4. Nemotron-Nano-30B: 444,544 samples
5. GLM-4.5-Air: 225,179 samples
6. MiniMax-M2.1: 204,773 samples
7. Baichuan-M3-235B: 124,520 samples
TOTAL BEFORE DEDUPLICATION: 2,919,699 samples
TOKEN COUNTS:
- Content tokens: 2.22 Billion
- Reasoning tokens: 1.56 Billion
- Total tokens: 3.78 Billion
- Samples with chain-of-thought: 100%
Quick Start:
from datasets import load_dataset
ds = load_dataset("OpenMed/Medical-Reasoning-SFT-Mega")All datasets Apache 2.0 licensed. Free for research and commercial use.
Thank you for following OpenMed's release series. I can't wait to see what you build. 🔥
OpenMed/Medical-Reasoning-SFT-Mega
OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B-V2
OpenMed/Medical-Reasoning-SFT-Trinity-Mini
OpenMed/Medical-Reasoning-SFT-GLM_4.5_Air
OpenMed/Medical-Reasoning-SFT-MiniMax-M2.1
OpenMed/Medical-Reasoning-SFT-Qwen3-Next-80B
OpenMed/Medical-Reasoning-SFT-Nemotron-Nano-30B
https://huggingface.co/datasets/OpenMed/Medical-Reasonin
https://huggingface.co/collections/OpenMed/medical-datasets
-
-
- 6 replies
danielhanchen
posted an
update 18 days ago
Post
3671
We created a tool-calling guide for local LLMs!
Learn how to use any open model like Qwen3-Coder-Next and GLM-4.7-Flash for function calling.
Guide: https://unsloth.ai/docs/basics/tool-calling-guide-for-local-llms
We provide hands-on examples for: story writing, Python execution, terminal tool calls, maths and more.
Learn how to use any open model like Qwen3-Coder-Next and GLM-4.7-Flash for function calling.
Guide: https://unsloth.ai/docs/basics/tool-calling-guide-for-local-llms
We provide hands-on examples for: story writing, Python execution, terminal tool calls, maths and more.
-
- 7 replies
danielhanchen
posted an
update 20 days ago
Post
3744
Qwen releases Qwen3-Coder-Next! 💜 Run the locally on 46GB RAM or less.
Thhe model excels at agentic coding & local use. With 256K context, it delivers similar performance to models with 10-20× more active parameters.
GGUF: unsloth/Qwen3-Coder-Next-GGUF
Guide: https://unsloth.ai/docs/models/qwen3-coder-next
Thhe model excels at agentic coding & local use. With 256K context, it delivers similar performance to models with 10-20× more active parameters.
GGUF: unsloth/Qwen3-Coder-Next-GGUF
Guide: https://unsloth.ai/docs/models/qwen3-coder-next
-
-
- 10 replies
danielhanchen
posted an
update 26 days ago
Post
3428
You can now run Kimi K2.5 locally! 🔥
We shrank the 1T model to 240GB (-60%) via Dynamic 1-bit.
Get >40 tok/s on 242GB or 622GB VRAM/RAM for near full precision.
GGUF: unsloth/Kimi-K2.5-GGUF
Guide: https://unsloth.ai/docs/models/kimi-k2.5
We shrank the 1T model to 240GB (-60%) via Dynamic 1-bit.
Get >40 tok/s on 242GB or 622GB VRAM/RAM for near full precision.
GGUF: unsloth/Kimi-K2.5-GGUF
Guide: https://unsloth.ai/docs/models/kimi-k2.5
-
- 7 replies
danielhanchen
posted an
update about 1 month ago
Post
2615
You can now fine-tune embedding models in our free Unsloth notebook! 🤗
Fine-tuning embedding models improves retrieval & RAG by aligning vectors to your domain-specific notion of similarity, improving search, clustering, and recommendations on your data.
⭐ Blog + Notebooks: https://unsloth.ai/docs/new/embedding-finetuning
Unsloth trains embedding models 1.8-3.3x faster with 20% less VRAM, 2x longer context & no accuracy loss vs. FA2 setups.
We'd like to thank Hugging Face and Unsloth contributor: electroglyph for making this possible!
Fine-tuning embedding models improves retrieval & RAG by aligning vectors to your domain-specific notion of similarity, improving search, clustering, and recommendations on your data.
⭐ Blog + Notebooks: https://unsloth.ai/docs/new/embedding-finetuning
Unsloth trains embedding models 1.8-3.3x faster with 20% less VRAM, 2x longer context & no accuracy loss vs. FA2 setups.
We'd like to thank Hugging Face and Unsloth contributor: electroglyph for making this possible!
-
- 3 replies
danielhanchen
posted an
update about 1 month ago
Post
2627
Run GLM-4.7-Flash locally on your device with 24GB RAM!🔥
It's the best performing 30B model on SWE-Bench and GPQA. With 200K context, it excels at coding, agents, chat & reasoning.
GGUF: unsloth/GLM-4.7-Flash-GGUF
Guide: https://unsloth.ai/docs/models/glm-4.7-flash
It's the best performing 30B model on SWE-Bench and GPQA. With 200K context, it excels at coding, agents, chat & reasoning.
GGUF: unsloth/GLM-4.7-Flash-GGUF
Guide: https://unsloth.ai/docs/models/glm-4.7-flash