Skip to content

shuu05/new_rep

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

19 Commits
 
 
 
 

Repository files navigation

SLM Model Catalog (7B-Class Focus)

SLM catalog covers ~7B parameter open-weight models (or small variants), highlighting key factors like permissive licenses, active communities, and low incident rates. Data draws from prior incident/literature reports; commercial use often allowed but with attribution; most accessible via Hugging Face.

Model Model Card Summary License Literature (Key Dates) Incidents (Key Dates) Commercial Restrictions Attribution Community Activity Accessibility
Llama 3 (Meta, Apr 2024) 8B base/instruct; multilingual; strong coding/math. Key: Safety-tuned via Guard. Llama 3 Community License (permissive, research+commercial) "Llama 3 Herd" paper (Jul 2024) [11]; Lakera risk report (May 2025) [12] Jailbreak (Apr 2024) [13]; CVE-2024-50050 RCE (Sep-Oct 2024) [14] None major; accept terms "Powered by Llama" optional High: 100M+ HF downloads; Ollama integration HF: meta-llama/Llama-3-8B; Docs
Llama 4 (Meta, Apr 2025) Scout/Maverick (MoE multimodal); 10M context. Key: Frontier-level scale. Llama 4 Community License "Llama 4 Challenges" (Apr 2025) [15]; Multimodal AI paper (Apr 2025) [16] Jailbreak 67% ASR Scout (Jul 2025) [17]; CVE-2025-53002 (Jun 2025) [18] Rate limits for large-scale Recommended badge Very High: Rapid adoption post-launch HF: meta-llama/Llama-4-Scout; Site
Code Llama (Meta, Aug 2023) 7/13/34/70B code-specialized from Llama 2. Key: Fill-mask coding. Same as Llama 2 Community PatchLM fixes paper (2025) [19]; Vuln detection (Sep 2025) [20] Framework RCE (Jan 2025) [21]; no model-specific None "Built with Code Llama" High: GitHub coding leader HF: codellama/CodeLlama-7b; Docs
Mistral 7B (Mistral AI, Sep 2023) Efficient dense; v0.3 sliding window. Key: Tops open LLM Arena. Apache 2.0 (most permissive) Mistral frontier news (Sep 2023) [22] CSAM gen in Pixtral (Jul 2025) [23]; policy update (Dec 2024) [24] None None required Very High: Ollama default; 50M+ downloads HF: mistralai/Mistral-7B-v0.3; Docs
MPT-7B (Databricks, May 2023) 1T tokens trained; 65k context ALiBi. Key: Early long-context. Apache 2.0 Databricks blog (Nov 2025 update) [25]; CyberLLM safety (Aug 2024) [26] None specific; OWASP general (2023+) [27] None None Medium: Legacy but GGUF active HF: mosaicml/mpt-7b; Blog
Gemma-7B (Google, Feb 2024) Gemini-derived; safety classifiers. Key: Lightweight rival to 70B. Gemma License (commercial ok) Tech report (Mar 2024) [28]; Cyber incidents (Apr 2025) [29] Gemini Trifecta indirect (Oct 2025) [30]; no direct High-stakes review advised "Made with Gemma" High: Google ecosystem push HF: google/gemma-7b [6]; Docs
Phi-3 Mini (Microsoft, Apr 2024) 3.8B SLM; synthetic data tuned. Key: Edge-first safety. MIT (highly permissive) Tech report (Apr 2024) [31]; Azure safety (Jan 2025) [32] None reported; strong RLHF [33] Production safeguards urged None High: Azure/ONNX optimized HF: microsoft/Phi-3-mini; Docs
StableLM 7B (Stability AI, Apr 2023) Early open chat-tuned. Key: Community fine-tunes. Stability AI License (CC-BY-SA-like) Safety Index evals (Winter 2025) [34] None specific; GenAI patterns [35] Non-commercial core? (check) Attribution required Medium: Zephyr variant active HF: stabilityai/stablelm-7b; Docs
Qwen 7B (Alibaba, Apr 2023+) Multilingual MoE; math/coding strong. Key: Chinese-English balance. Apache 2.0 Qwen2 tech report (Jun 2024) Low jailbreak resistance (2024 evals) None Optional High: Asia dev focus HF: Qwen/Qwen-7B; Site
StarCoder/Base (BigCode, May 2023) 15B code (use 7B equiv); 80+ langs. Key: Permissive code data. BigCode OpenRAIL-M (restrict harmful) StarCoder paper (Oct 2023) Code vuln gen risks (2024) No illegal code use Cite BigCode High: Coding hub HF: bigcode/starcoder; Docs
CodeGen-2 (Salesforce, Aug 2022) Multi-turn code; 7B/16B. Key: Early infill. Salesforce License (research focus) CodeGen paper (Mar 2022) General code risks Commercial limited Attribution Low-Medium: Legacy HF: Salesforce/codegen-7B
CodeParrot (HF, Mar 2022) GitHub-trained code. Key: Repo-level. Apache 2.0 CodeParrot paper (2022) Vuln insertion (2023 studies) None None Low: Superseded HF: codeparrot/codeparrot
RWKV 7B (RWKV, 2023 variants) RNN-LM hybrid; infinite ctx. Key: GPU-efficient. Apache 2.0 RWKV v4 paper (2023); v6 (2024) Few; stable RNN arch None Cite RWKV Medium: RNN niche HF: RWKV/rwkv-7b; Docs
Falcon-7B (TII, May 2023) RefinedWeb trained. Key: Early SOTA open. Apache 2.0 Falcon paper (May 2023) Early jailbreaks (2023) None None Medium: Chat variants HF: tiiuae/falcon-7b; Site
RedPajama-7B (Together, Mar 2023) LLaMA dupereplica; 1T tokens. Key: Open data recipe. Apache 2.INC RedPajama paper (Apr 2023) Data poisoning risks None Cite dataset Medium: Base for fine-tunes HF: togethercomputer/RedPajama

Key Highlights

  • Safest for Commercial: Mistral/Phi-3/Qwen (Apache/MIT; zero restrictions).
  • Highest Incidents: Llama series (framework CVEs, jailbreaks); code models risk vuln gen.
  • Most Active Communities: Llama/Mistral/Gemma (HF stars >1M, Ollama defaults).
  • Edge-Friendly: Phi-3/Gemma (quantized, <5GB RAM).
  • Download Tip: All primary on Hugging Face; use ollama pull for local (e.g., ollama pull mistral:7b). Check licenses for derivatives.[3]

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 2

  •  
  •