base_model: usermma/tinybrainbot-53m-a9m-hybrid-instruct-abliterated
tags:
- gguf
- lfm2-moe
- abliterated
- tiny
license: apache-2.0
tinybrainbot-53m-a9m-hybrid-instruct-abliterated — GGUF
GGUF conversions of usermma/tinybrainbot-53m-a9m-hybrid-instruct-abliterated
(LFM2-MoE architecture, 53M total / ~9M active params, 32 experts, top-4 routing).
Files
| File | Quantization | Size |
|---|---|---|
tinybrainbot-53m-a9m-hybrid-instruct-abliterated.F16.gguf |
16-bit float | ~104 MB |
tinybrainbot-53m-a9m-hybrid-instruct-abliterated.Q8_0.gguf |
8-bit | ~56 MB |
tinybrainbot-53m-a9m-hybrid-instruct-abliterated.Q6_K.gguf |
6-bit | ~56 MB |
tinybrainbot-53m-a9m-hybrid-instruct-abliterated.Q4_K_M.gguf |
4-bit | ~45 MB |
tinybrainbot-53m-a9m-hybrid-instruct-abliterated.Q2_K.gguf |
2-bit | ~38 MB |
Details
- Converted with llama.cpp
convert_hf_to_gguf.py(LFM2-MoE support) +llama-quantize - Vocabulary: 49,152-token byte-level BPE (qwen2-style pre-tokenizer, NFC normalization)
- Chat template embedded (
tokenizer.chat_template) - Conversation format: ChatML-style with
<|im_start|>/<|im_end|>and thinking tags
Usage
llama-cli -m tinybrainbot-53m-a9m-hybrid-instruct-abliterated.Q4_K_M.gguf -p "Hello!" -n 64