cenkierpiotr/MEMORYDistillPL-1.5b

TEXT GENERATIONPricing:Input $0.04 / Cached $0.008 / Output $0.08Concurrent Unit Cost:1Model Size:1.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Sep 29, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

MEMORYDistillPL-1.5b by cenkierpiotr is a 1.5 billion parameter Qwen2.5-1.5B-Instruct fine-tune, specifically optimized for distilling Polish-language conversational transcripts into structured memory notes and extracting entities. This model features a 32,768 token context window and is designed as a lightweight, fast alternative for this narrow summarization and extraction task, outperforming larger general-purpose LLMs in its specialized domain.

Loading preview...

MEMORYDistillPL-1.5b: Specialized Polish Conversation Distillation

MEMORYDistillPL-1.5b, developed by cenkierpiotr, is a 1.5 billion parameter model fine-tuned from Qwen2.5-1.5B-Instruct using QLoRA. Its core purpose is to distill long Polish-language conversational transcripts into concise, structured memory notes, summarizing discussions and extracting key entities like people, organizations, decisions, and facts. This model is highly optimized for this specific task, rather than general-purpose chat, offering a lightweight and fast solution.

Key Capabilities

  • Structured Summarization: Transforms Polish conversation transcripts into structured, factual summaries.
  • Entity Extraction: Identifies and extracts entities (people, organizations, decisions, facts) from conversations.
  • High Context Window: Supports a 32,768 token context, allowing for processing of lengthy transcripts.
  • Performance: In blind pairwise evaluations, it was preferred in approximately 67% of comparisons against the 4.5x larger Qwen2.5-7B-Instruct for its intended task.

Good for

  • Distilling Polish-language conversation transcripts into structured, factual summaries.
  • Extracting specific entities from Polish conversational text.
  • Use cases requiring a lightweight and fast alternative to larger general-purpose LLMs for specialized summarization/extraction in Polish.

Limitations

  • Not intended for general open-domain chat or instruction-following outside its specialized task.
  • May occasionally hallucinate plausible but incorrect English expansions for Polish abbreviations.
  • Can sometimes miss entities, requiring human verification for high-stakes applications.

Finetunes
1 models