Agaram 2.5B (அகரம்) — base model GGUF
GGUF files of the Agaram 2.5B base model for llama.cpp and tools built on it. Agaram is a Tamil–English model trained from scratch by Murai Labs and Ramchand Kumaresan; it is not a fine-tune of any other model. Model card: base. Code: Murai-Labs/Agaram2.5b. Paper: arXiv:2610.18420.
This is a base model: it continues text and does not follow instructions or chat. To chat, use
murai-labs/agaram-2.5b-gguf (for example
ollama run hf.co/murai-labs/agaram-2.5b-gguf:Q8_0), or try the
chat demo.
These files were built on 2026-10-12 with llama.cpp b11574 and were in murai-labs/agaram-2.5b-gguf until
2026-10-29; the files and hashes are unchanged. File hashes are in SHA256SUMS.
| File | Size | Use |
|---|---|---|
agaram-2.5b-f16.gguf |
5.06 GB | base, half precision |
agaram-2.5b-Q8_0.gguf |
2.69 GB | base, 8-bit |
agaram-2.5b-Q4_K_M.gguf |
1.56 GB | base, small 4-bit |
llama-completion -m agaram-2.5b-Q8_0.gguf -p "சென்னை ஒரு" -n 128 --temp 0
- The files stop on
<|endoftext|>(id 0). - Context: 8,192 tokens.
Requirements: a llama.cpp that knows the qwen35 pre-tokenizer
Agaram's tokenizer is a byte-level BPE whose pre-tokenizer is Qwen2's split pattern with Unicode combining marks and
the zero-width joiners added to the letter class, so that Tamil vowel signs stay attached to their consonants.
llama.cpp has no entry for this exact pattern. These files use llama.cpp's qwen35 pre-tokenizer, whose pattern is the
same except that it does not include the zero-width joiners (U+200C, U+200D). Use a llama.cpp build that recognises
qwen35 (b11574 or newer); older builds, and apps that bundle them, will refuse to load the files or tokenize
differently.
What was checked
- Tokenization (f16 file, 2026-10-12). llama.cpp's token ids were compared with the Hugging Face tokenizer on 961 texts: every prompt and reference answer of our three chat evaluation sets (661 texts in Tamil script, Tanglish and English) and 300 held-out training conversations rendered with the chat template. All 961 match exactly. The Q8_0 and Q4_K_M base files were not tested separately. None of these texts contains a zero-width joiner, the one case where the patterns differ, so that case is untested.
Model, evaluation, limitations and disclosures
See the base model card.
Creators
Murai Labs and Ramchand Kumaresan. Licence: Apache-2.0; upstream data terms in the base model card. Technical report: https://arxiv.org/abs/2610.18420