`, use `
`/`
` for sections in long text 3. Lists with `
- `, quotes with `
`, code with `
`, important sentences can use ``
4. Preserve all original information, no additions, deletions, or modifications
5. No ``/``/`` document-level tags
Let me read the text carefully:
The text is about Romanized Nepali language models. It's an academic abstract covering:
- Background: Romanized Nepali in digital communication, no standard spelling, unclear how reliable small open language models can be adapted
- Method: Adapted three 2-3B base models (Llama-3.2-3B, Qwen2.5-3B, Gemma-2-2B) with QLoRA on 9,000 instruction examples
- Evaluation: 1,000 held-out examples with metrics, compared against zero-shot baselines
- Results: Fine-tuning raises chrF++ from 10-14 to 29.0-30.6, lowers BPB from 2.23-2.41 to 0.78-0.84
- Detailed results for each model
- Supplementary comparison with Llama-3.1-8B
- Conclusion: Lower fertility and higher reference likelihood did not identify the model whose outputs were closest to references
Now I need to structure this as semantic HTML. Since it's an academic abstract, I should:
- Use `` for main sections (like "Background", "Methodology", "Results", "Conclusion" - but wait, rule 4 says don't add or change content. Can I add section headings that weren't in the original? The rule says "不得增删改内容" (must not add, delete, or modify content). Hmm, but rule 2 says "长文用
/
划分小节" (use h2/h3 to divide sections in long text). So I can add structural headings, but I should use the original text's own structure. Since the original doesn't have explicit section headings, I should be careful. Maybe I can use minimal headings that reflect the content structure, or just use paragraphs. Actually, the instruction says to use h2/h3 for dividing sections in long text - this is about formatting structure, not adding new content. I think adding section headings that summarize parts of the text could be seen as adding content. Let me think...
The safest approach: use `
` for each natural paragraph. The original text appears to be one continuous abstract. I could split it into logical paragraphs based on topic shifts. But rule 4 says preserve all original information without adding. Adding section headings like "Background" would be adding labels that weren't in the original. However, rule 2 explicitly says to use h2/h3 for sections in long text. This is a formatting directive. I think using h2/h3 for major topic shifts is acceptable as layout, but to be safe, I could just structure it as paragraphs since the original is a single abstract without clear sections.
Actually, let me re-read: "长文用
`, code with ``, important sentences can use `` 4. Preserve all original information, no additions, deletions, or modifications 5. No ``/``/`` document-level tags Let me read the text carefully: The text is about Romanized Nepali language models. It's an academic abstract covering: - Background: Romanized Nepali in digital communication, no standard spelling, unclear how reliable small open language models can be adapted - Method: Adapted three 2-3B base models (Llama-3.2-3B, Qwen2.5-3B, Gemma-2-2B) with QLoRA on 9,000 instruction examples - Evaluation: 1,000 held-out examples with metrics, compared against zero-shot baselines - Results: Fine-tuning raises chrF++ from 10-14 to 29.0-30.6, lowers BPB from 2.23-2.41 to 0.78-0.84 - Detailed results for each model - Supplementary comparison with Llama-3.1-8B - Conclusion: Lower fertility and higher reference likelihood did not identify the model whose outputs were closest to references Now I need to structure this as semantic HTML. Since it's an academic abstract, I should: - Use `` for main sections (like "Background", "Methodology", "Results", "Conclusion" - but wait, rule 4 says don't add or change content. Can I add section headings that weren't in the original? The rule says "不得增删改内容" (must not add, delete, or modify content). Hmm, but rule 2 says "长文用
/
划分小节" (use h2/h3 to divide sections in long text). So I can add structural headings, but I should use the original text's own structure. Since the original doesn't have explicit section headings, I should be careful. Maybe I can use minimal headings that reflect the content structure, or just use paragraphs. Actually, the instruction says to use h2/h3 for dividing sections in long text - this is about formatting structure, not adding new content. I think adding section headings that summarize parts of the text could be seen as adding content. Let me think... The safest approach: use `
` for each natural paragraph. The original text appears to be one continuous abstract. I could split it into logical paragraphs based on topic shifts. But rule 4 says preserve all original information without adding. Adding section headings like "Background" would be adding labels that weren't in the original. However, rule 2 explicitly says to use h2/h3 for sections in long text. This is a formatting directive. I think using h2/h3 for major topic shifts is acceptable as layout, but to be safe, I could just structure it as paragraphs since the original is a single abstract without clear sections. Actually, let me re-read: "长文用
/
划分小节" - "for long text, use h2/h3 to divide into sections". The text is about 250 words, which is a medium-length abstract. I think I can either treat it as one section with paragraphs, or add a few section headings. Given the ambiguity and the strict "no additions" rule, I'll lean toward using mostly paragraphs, but perhaps one h2 for the overall topic... Actually no. Let me just use paragraphs since the original doesn't have explicit sections. Or maybe I can use h2 to introduce the study.
Hmm, let me think about what makes good semantic HTML for an abstract:
- Could use h2 for "Background" or similar
- But since the original is one continuous abstract, splitting it might make sense
Actually, many academic abstracts have a "Keywords" section, but this one doesn't. Let me just format it as a series of paragraphs, breaking at logical points.
Let me structure the content:
Paragraph 1: Background/motivation: "Romanized Nepali, Nepali written in Latin script, is common in digital communication but has no standard spelling and few resources, and it is unclear how reliably small open language models can be adapted to it."
Paragraph 2: Method: "We adapt three 2-3B base models from different families, Llama-3.2-3B, Qwen2.5-3B, and Gemma-2-2B, with QLoRA on 9,000 instruction examples produced by deterministic transliteration of Nepali instruction data."
Paragraph 3: Evaluation setup: "Each model is trained with one shared recipe and three seeds and evaluated on 1,000 held-out examples with reference-overlap metrics, byte-normalized reference likelihood (bits per byte, BPB), and an output audit, against zero-shot baselines under two prompts."
Paragraph 4: Main results: "Fine-tuning raises chrF++ from 10-14 to 29.0-30.6 and lowers BPB from 2.23-2.41 to 0.78-0.84; most zero-shot outputs (54-92%) are English-like."
Paragraph 5: Model comparison: "Likelihood and overlap disagree on how far apart the models are."
Paragraph 6: Llama vs Qwen: "Llama-3.2-3B has the lowest BPB in every seed, 0.042 below Qwen2.5-3B (95% CI 0.039-0.046), yet the two cannot be separated on chrF++ (30.51 and 30.61) or BLEU, even with paired item-level tests."
Paragraph 7: Gemma results: "Gemma-2-2B, whose tokenizer splits Romanized Nepali words least (2.72 tokens per word), scores worst on every measure."
Paragraph 8: Prompt dependence: "Which model gains most over its zero-shot baseline depends on the zero-shot prompt."
Paragraph 9: Supplementary comparison: "In a supplementary comparison with an identical tokenizer, Llama-3.1-8B lowers BPB by 0.053 (0.050-0.056) relative to Llama-3.2-3B, while chrF++ and BLEU rise by only 0.68 and 0.58 points."
Paragraph 10: Conclusion: "Lower fertility and higher reference likelihood therefore did not identify the model whose outputs were closest to the references."
Model names could be in `` or kept as plain text. Metrics like chrF++ and BPB could be in `` as they're key results. Numbers like model sizes could be kept as-is.
Should I use a model list as `
`? The three models could be listed, but in the original they're inline in a sentence. Rule 4 says don't change content. I'll keep them inline. Actually, wait - let me reconsider. Rule 3 says lists should be in ul/li. The original mentions "three 2-3B base models from different families" and names them inline. If I convert this to a list, that changes the structure but not the information. Hmm, "不得增删改内容" - adding a list structure changes the presentation but the content (the three model names) remains. I think it's safer to keep the sentence intact as the original flows naturally. But the instruction #3 says "列表内容用
- " - if there's list content, use ul/li. The original isn't really list content; it's a sentence. I'll keep it as a sentence. For emphasis: I can use `` on key findings: - "no standard spelling and few resources" - The metric improvements - "Lower fertility and higher reference likelihood therefore did not identify..." (conclusion) Let me be moderate with strong emphasis to keep
- " - if there's list content, use ul/li. The original isn't really list content; it's a sentence. I'll keep it as a sentence. For emphasis: I can use `` on key findings: - "no standard spelling and few resources" - The metric improvements - "Lower fertility and higher reference likelihood therefore did not identify..." (conclusion) Let me be moderate with strong emphasis to keep