DAC5-3B / README.md
Mattimax's picture
Update README.md
ed67744 verified
metadata
license: mit
datasets:
  - Mattimax/DACMini_Refined
  - Mattimax/Camoscio-ITA
language:
  - it
  - en
library_name: transformers
tags:
  - DAC
  - M.INC.
  - conversational

โ˜• Support the project

Buy Me a Coffee

๐Ÿ‡ฎ๐Ÿ‡น ITALIANO

๐Ÿ“˜ Model Card โ€” Mattimax/DAC5-3B

๐Ÿง  Informazioni Generali

  • Nome: Mattimax/DAC5-3B
  • Serie: DAC (DATA-AI Chat) โ€“ 5ยช versione
  • Autore: Mattimax
  • Research Lab / Azienda: MINC01
  • Base Model: Qwen โ€“ Qwen2.5-3B-Instruct

DAC5-3B รจ attualmente il modello piรน avanzato e sperimentale della serie DAC, progettato per massimizzare qualitร  conversazionale, integrandosi al meglio con server MCP e performance tecnica su architettura 3B.


๐Ÿ— Architettura Tecnica

Core Architecture

  • Architettura: Qwen2ForCausalLM
  • Parametri: ~3B
  • Numero layer: 36
  • Hidden size: 2048
  • Intermediate size: 11008
  • Attention heads: 16
  • Key/Value heads (GQA): 2
  • Attivazione: SiLU
  • Norm: RMSNorm (eps 1e-6)
  • Tie word embeddings: Yes

Attention

  • Full attention su tutti i 36 layer
  • Attention dropout: 0.0
  • Sliding window: disabilitato
  • GQA (Grouped Query Attention) โ†’ maggiore efficienza memoria

Positional Encoding

  • Max position embeddings: 32768
  • RoPE theta: 1,000,000
  • RoPE scaling: None

Precision & Performance

  • Torch dtype: bfloat16
  • Quantizzazione training: 4-bit (NF4)
  • Cache abilitata per inference
  • Ottimizzato con Unsloth (fixed build 2026.2.1)

Tokenizer

  • Vocab size: 151,936
  • EOS token id: 151645
  • PAD token id: 151654

๐ŸŽฏ Obiettivo del Modello

DAC5-3B รจ stato progettato per:

  • ๐Ÿ‡ฎ๐Ÿ‡น Massima qualitร  in italiano
  • โšก Alta efficienza su GPU consumer
  • ๐Ÿงฉ Conversazione coerente multi-turn
  • ๐Ÿ› ๏ธ Supporto tecnico e coding leggero
  • ๐Ÿง  Migliore stabilitร  rispetto ai DAC precedenti

รˆ un modello orientato a sviluppatori indipendenti, maker e sistemi offline (come OpenClaw, Claude Code, OpenCode, ecc...)


๐Ÿ“š Dataset & Specializzazione

Il fine-tuning supervisionato รจ stato effettuato su un mix altamente selezionato di dataset italiani:

  • Camoscio-ITA
  • DACMini Refined
  • Conversazioni sintetiche italiane ad alta qualitร 

Strategia

  • Dataset limitato ma ad alta densitร  informativa (~20k esempi)
  • Minimizzazione del rumore
  • Focus su chiarezza e coerenza
  • Riduzione delle risposte generiche tipiche dei 3B

๐Ÿš€ Capacitร  Principali

DAC5-3B eccelle in:

  • Spiegazioni tecniche
  • Scrittura strutturata
  • Programmazione livello medio
  • Traduzione IT โ†” EN
  • Brainstorming progettuale
  • Assistenti locali offline
  • Supporto allo studio

๐Ÿ“Š Differenze rispetto ai DAC precedenti

โœ” Maggiore stabilitร  nelle risposte lunghe โœ” Meno ripetizioni โœ” Migliore controllo del tono โœ” Risposte piรน dirette โœ” Migliore allineamento alle istruzioni

DAC5 rappresenta il punto piรน alto raggiunto finora nella serie.


โš ๏ธ Limitazioni

  • Contesto di training effettivo: 1024 token
  • Non ottimizzato per tool calling complesso
  • Non specializzato in matematica avanzata
  • Puรฒ degradare su reasoning multi-step molto profondo
  • Modello sperimentale

๐Ÿ’ป Requisiti Hardware

Inference consigliata

  • GPU 6โ€“8GB VRAM (quantizzato)
  • Oppure CPU moderna con GGUF

Compatibile con:

  • PC consumer
  • Mini workstation
  • Sistemi edge
  • Setup locali offline

๐Ÿ”ฌ Filosofia DAC

La serie DAC nasce con l'obiettivo di:

Spingere al massimo modelli compatti, ottimizzando qualitร  reale invece di scalare solo i parametri.

DAC5-3B รจ il risultato piรน maturo di questa filosofia: qualitร  elevata su architettura 3B con risorse contenute.


๐Ÿงช Stato del Modello

๐ŸŸก Sperimentale ma stabile รˆ il miglior modello della serie DAC fino ad oggi, ma rimane parte di un ciclo evolutivo continuo.


๐Ÿ“š Citation

Se utilizzi Mattimax/DAC5-3B nei tuoi lavori di ricerca, progetti o pubblicazioni, puoi citarlo nel seguente modo:

@misc{mattimax_dac5_3b_2026,
  author       = {Mattimax},
  title        = {DAC5-3B: Fifth Iteration of the Dynamic Adaptive Core Series},
  year         = {2026},
  publisher    = {Hugging Face},
  organization = {MINC01},
  note         = {Experimental Italian-specialized 3B language model},
  url          = {https://huggingface.co/Mattimax/DAC5-3B}
}

Citazione testuale breve:

Mattimax. DAC5-3B: Fifth Iteration of the Dynamic Adaptive Core Series. 2026. MINC01 Research Lab.



๐Ÿ‡ฌ๐Ÿ‡ง ENGLISH

๐Ÿ“˜ Model Card โ€” Mattimax/DAC5-3B

๐Ÿง  General Information

  • Name: Mattimax/DAC5-3B
  • Series: DAC (DATA-AI Chat) โ€“ 5th version
  • Author: Mattimax
  • Research Lab / Company: MINC01
  • Base Model: Qwen โ€“ Qwen2.5-3B-Instruct

DAC5-3B is currently the most advanced and experimental model in the DAC series, designed to maximize conversational quality, integrating better with MCP servers and technical performance on a 3B architecture.


๐Ÿ— Technical Architecture

Core Architecture

  • Architecture: Qwen2ForCausalLM
  • Parameters: ~3B
  • Number of layers: 36
  • Hidden size: 2048
  • Intermediate size: 11008
  • Attention heads: 16
  • Key/Value heads (GQA): 2
  • Activation: SiLU
  • Norm: RMSNorm (eps 1e-6)
  • Tie word embeddings: Yes

Attention

  • Full attention across all 36 layers
  • Attention dropout: 0.0
  • Sliding window: disabled
  • GQA (Grouped Query Attention) โ†’ improved memory efficiency

Positional Encoding

  • Max position embeddings: 32768
  • RoPE theta: 1,000,000
  • RoPE scaling: None

Precision & Performance

  • Torch dtype: bfloat16
  • Training quantization: 4-bit (NF4)
  • Cache enabled for inference
  • Optimized with Unsloth (fixed build 2026.2.1)

Tokenizer

  • Vocab size: 151,936
  • EOS token id: 151645
  • PAD token id: 151654

๐ŸŽฏ Model Objective

DAC5-3B was designed for:

  • ๐Ÿ‡ฎ๐Ÿ‡น Maximum Italian language quality
  • โšก High efficiency on consumer GPUs
  • ๐Ÿงฉ Coherent multi-turn conversations
  • ๐Ÿ› ๏ธ Technical support and light coding
  • ๐Ÿง  Improved stability compared to previous DAC versions

It is oriented toward independent developers, makers, and offline systems (such as OpenClaw, Claude Code, OpenCode, etc.).


๐Ÿ“š Dataset & Specialization

Supervised fine-tuning was performed on a highly curated mix of Italian datasets:

  • Camoscio-ITA
  • DACMini Refined
  • High-quality synthetic Italian conversations

Strategy

  • Limited but high-density dataset (~20k samples)
  • Noise minimization
  • Focus on clarity and coherence
  • Reduction of generic 3B-style responses

๐Ÿš€ Core Capabilities

DAC5-3B excels at:

  • Technical explanations
  • Structured writing
  • Intermediate-level programming
  • IT โ†” EN translation
  • Project brainstorming
  • Offline local assistants
  • Study support

๐Ÿ“Š Differences from Previous DAC Versions

โœ” Greater stability in long responses โœ” Fewer repetitions โœ” Better tone control โœ” More direct answers โœ” Improved instruction alignment

DAC5 represents the highest point reached so far in the series.


โš ๏ธ Limitations

  • Effective training context: 1024 tokens
  • Not optimized for advanced tool calling
  • Not specialized in advanced mathematics
  • May degrade in very deep multi-step reasoning
  • Experimental model

๐Ÿ’ป Hardware Requirements

Recommended Inference

  • 6โ€“8GB VRAM GPU (quantized)
  • Or modern CPU with GGUF

Compatible with:

  • Consumer PCs
  • Mini workstations
  • Edge systems
  • Offline local setups

๐Ÿ”ฌ DAC Philosophy

The DAC series was created with the goal of:

Pushing compact models to their limits, optimizing real quality instead of merely scaling parameters.

DAC5-3B is the most mature result of this philosophy: high quality on a 3B architecture with limited resources.


๐Ÿงช Model Status

๐ŸŸก Experimental but stable It is the best model in the DAC series to date, but remains part of an ongoing evolutionary cycle.


๐Ÿ“š Citation

If you use Mattimax/DAC5-3B in research work, projects, or publications, you may cite it as follows:

@misc{mattimax_dac5_3b_2026,
  author       = {Mattimax},
  title        = {DAC5-3B: Fifth Iteration of the Dynamic Adaptive Core Series},
  year         = {2026},
  publisher    = {Hugging Face},
  organization = {MINC01},
  note         = {Experimental Italian-specialized 3B language model},
  url          = {https://huggingface.co/Mattimax/DAC5-3B}
}

Short textual citation:

Mattimax. DAC5-3B: Fifth Iteration of the Dynamic Adaptive Core Series. 2026. MINC01 Research Lab.