GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
sinhala-nlp
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
sinhala-nlp
Model Tree
Qwen/Qwen3.5-4B
Input Modalities
Output Modalities
Supported Functionality
python
from transformers import AutoTokenizer, AutoModelForCausalLMfrom peft import PeftModeltok = AutoTokenizer.from_pretrained("Qwen/Qwen3.5-4B")model = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-4B", dtype="auto", device_map="auto")model = PeftModel.from_pretrained(model, "sinhala-nlp/Qwen3.5-4B-PaliSinhala-Pali2Si-en")
Prompts use the base model's chat template with thinking disabled, and the assistant
response begins with the Translation: prefix.
| Training pairs | 9450 |
| Instruction language | en |
| Length-matched subsample | False |
| Epochs | 1.0 |
| Effective batch size | 16 |
| Learning rate | 0.0002 |
| Max sequence length | 3072 |
| LoRA r / alpha / dropout | 16 / 32 / 0.05 |
| Thinking during training | False |
Last 1000 rows of sinhala-nlp/pali-sinhala, whitespace-tokenized (sacreBLEU's default
13a tokenizer splits Sinhala conjuncts and vowel signs):
| Metric | Score |
|---|---|
| Corpus sacreBLEU | 0.06 |
| Sentence-level BLEU mean | 0.21 |
| Hyp/ref length ratio | 0.247 |
Read the scores with the length ratio. The corpus is in canonical order, so the trailing 1000 rows used as the test set are much longer than the training pool (median target length differs by roughly an order of magnitude). Predictions are therefore shorter than references and BLEU is partly driven by the brevity penalty rather than translation quality. This split is kept as-is so the numbers stay comparable across the model families evaluated in SinGen.
This adapter inherits the licence of the base model; check the base model card
before redistributing. Verify the terms of the sinhala-nlp/pali-sinhala dataset
on its dataset card as well.