DeepSeek-V4-Pro API Test Report: Together & Nvidia#4
Draft
yjfireworks wants to merge 2 commits intomainfrom
Draft
DeepSeek-V4-Pro API Test Report: Together & Nvidia#4yjfireworks wants to merge 2 commits intomainfrom
yjfireworks wants to merge 2 commits intomainfrom
Conversation
Test results for DSv4-Pro with high reasoning effort across 6 prompts: - Both providers return correct answers - Both exhibit gibberish/artifacts in reasoning traces - Nvidia has severe reliability issues (3/6 timeouts, high latency) - Together has token truncation risk with reasoning-heavy queries Co-authored-by: Yun Jin <yjfireworks@users.noreply.github.com>
12 parallel agents tested both APIs simultaneously: - Together: 12/12 success, 7/12 had reasoning gibberish, 1 leaked to content - Nvidia: only 1/6 succeeded (17%), 5/6 timed out at 200s - New artifact type: Nvidia leaked special token <end_of_repo_name> - Combined across both runs: Together 100% reliable, Nvidia 33% reliable Co-authored-by: Yun Jin <yjfireworks@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Test report evaluating DeepSeek-V4-Pro (DSv4-Pro) API responses from Together AI and Nvidia NIM with
reasoning_effort=high. Two runs were performed: Run 1 sequential (6 tests each), Run 2 fully parallel via 12 concurrent agents (6 tests each).Key Findings
Answer Correctness
Both providers produce correct final answers for all tested prompts when requests complete successfully. Tested across riddles, math, coding, and explanations.
Gibberish in Reasoning Traces (Both Providers)
The most significant finding — both providers exhibit artifacts/gibberish injected into reasoning traces:
"So771","the716716","we32","as42-is","But13;""key19275:","can16:","But71:","it790:""06"/"07"appearing 30-40 times in a single response"böjnings","dátummal","Didži""<｜end▁of▁repo▁name｜>"contentfield (Together, Stack vs Queue:"to400: different")Affected: 7/12 Together responses (58%), 4/4 successful Nvidia responses (100%).
Nvidia Reliability
Together Truncation
max_tokenslimit during reasoning, truncating contentCombined Results (24 total requests)
Slack Thread