dusersad12/SkylineLM-ReleaseRepo

SkylineLM

---

1. Introduction

The SkylineLM release in this repository is the outcome of a full re-run of our post-training pipeline with a larger reasoning-oriented mixture and a revised curriculum for the late training stages. It delivers stronger depth of reasoning and markedly better coding and general logic behaviour than any of our earlier checkpoints, while keeping the same inference footprint as the base model.

Compared with earlier internal builds, the reasoning behaviour of this release is noticeably more stable on long chains of thought. On a chain-of-thought arithmetic set the pass rate moved from 74% to 89% while average solution length grew from 9K to 21K tokens per problem, which lines up with what we see qualitatively: the model is willing to think longer before committing to an answer.

Alongside the reasoning gains, this release ships a lower hallucination rate on factual QA and a more reliable function-calling implementation.

2. Evaluation Results

Comprehensive Benchmark Results

	Benchmark	Helios-7B	Helios-13B	Helios-7B-v2	SkylineLM
Core Reasoning Tasks	Math Reasoning	0.482	0.501	0.495	0.558
	Logical Reasoning	0.745	0.768	0.779	0.785
	Common Sense	0.690	0.675	0.702	0.711
Language Understanding	Reading Comprehension	0.648	0.662	0.671	0.720
	Question Answering	0.559	0.577	0.583	0.611
	Text Classification	0.788	0.796	0.805	0.792
	Sentiment Analysis	0.754	0.762	0.771	0.782
Generation Tasks	Code Generation	0.592	0.608	0.617	0.611
	Creative Writing	0.564	0.553	0.576	0.559
	Dialogue Generation	0.598	0.612	0.616	0.663
	Summarization	0.719	0.731	0.738	0.692
Specialized Capabilities	Translation	0.756	0.773	0.778	0.797
	Knowledge Retrieval	0.628	0.644	0.649	0.657
	Instruction Following	0.708	0.724	0.728	0.727
	Safety Evaluation	0.695	0.679	0.703	0.772

Overall Performance Summary

The SkylineLM shows balanced strength across the whole suite, with the clearest headroom-to-margin gains on reasoning-heavy and coding tasks.

3. Chat Website & API Platform

We host a public chat UI and an inference API for SkylineLM; both are linked from the project website.

4. How to Run Locally

The model runs with any recent transformers installation. A minimal example, together with recommended sampling settings, lives in the code repository.

Two usage notes for this release:

1. A system prompt is supported and encouraged.
2. No special token is needed at the start of the output to trigger the thinking pattern.

The small variant of SkylineLM keeps the base architecture and shares the tokenizer of the main release, so it can be loaded exactly like its base model.

System Prompt

We recommend using the following system prompt with a specific date.

You are SkylineLM, a helpful AI assistant.
Today is {current date}.

For example,

You are SkylineLM, a helpful AI assistant.
Today is May 28, 2025, Monday.

Temperature

We recommend setting the temperature parameter $T_{model}$ to 0.6.

Prompts for File Uploading and Web Search

For file uploading, please follow the template to create prompts, where {file_name}, {file_content} and {question} are arguments.

file_template = \
"""[file name]: {file_name}
[file content begin]
{file_content}
[file content end]
{question}"""

For web search enhanced generation, we recommend the following prompt template where {search_results}, {cur_date}, and {question} are arguments.

search_answer_en_template = \
'''# The following contents are the search results related to the user's message:
{search_results}
In the search results I provide to you, each result is formatted as [webpage X begin]...[webpage X end], where X represents the numerical index of each article. Please cite the context at the end of the relevant sentence when appropriate. Use the citation format [citation:X] in the corresponding part of your answer. If a sentence is derived from multiple contexts, list all relevant citation numbers, such as [citation:3][citation:5]. Be sure not to cluster all citations at the end; instead, include them in the corresponding parts of the answer.
When responding, please keep the following points in mind:
- Today is {cur_date}.
- Not all content in the search results is closely related to the user's question. You need to evaluate and filter the search results based on the question.
- For listing-type questions (e.g., listing all flight information), try to limit the answer to 10 key points and inform the user that they can refer to the search sources for complete information. Prioritize providing the most complete and relevant items in the list. Avoid mentioning content not provided in the search results unless necessary.
- For creative tasks (e.g., writing an essay), ensure that references are cited within the body of the text, such as [citation:3][citation:5], rather than only at the end of the text. You need to interpret and summarize the user's requirements, choose an appropriate format, fully utilize the search results, extract key information, and generate an answer that is insightful, creative, and professional. Extend the length of your response as much as possible, addressing each point in detail and from multiple perspectives, ensuring the content is rich and thorough.
- If the response is lengthy, structure it well and summarize it in paragraphs. If a point-by-point format is needed, try to limit the answer to 5 points and merge related items.
- For objective Q&A, if the answer is very brief, you may add one or two related sentences to enrich the content.
- Choose an appropriate and visually appealing format for your response based on the user's requirements and the content of the answer, ensuring strong readability.
- Your answer should synthesize information from multiple relevant webpages and avoid repeatedly citing the same webpage.
- Unless the user requests otherwise, your response should be in the same language as the user's question.
# The user's message is:
{question}'''

5. License

This code repository is licensed under the MIT License. Use of the SkylineLM weights is also governed by the MIT License. Commercial use and distillation are both permitted.

6. Contact

If you have questions about the release, open an issue on the code repository or write to contact@skylinelm.ai.