Claude finished @jthomson04's task in 2m 1s —— View job

PR Review: [NV] DSR1 FP8 B300 Dynamo TRT

• Gather context and read changed files
• Validate master config + perf-changelog
• Validate container image accessibility
• Validate enroot import in launch script
• Validate model prefixes
• Check for bugs and logic errors
• Post review summary

Summary

This PR adds FP8 precision support for DSR1 on B300 with Dynamo TRT, including 27 benchmark configurations (1k1k and 8k1k, both MTP and STP variants), launch script updates to handle FP8 model paths, and a perf-changelog entry.

Validations passed:

• perf-changelog.yaml updated alongside nvidia-master.yaml ✓
• Image nvcr.io/nvidia/ai-dynamo/tensorrtllm-runtime:0.8.1.post1 is publicly accessible on NGC ✓
• enroot import docker://$IMAGE pattern present in launch script ✓
• Model prefix dsr1 is valid ✓
• Launch script FP8 handling mirrors the existing GB200 pattern (SRT_SLURM_MODEL_PREFIX, model path, served model name) ✓

One issue found — see inline comment:

• 🟡 All 27 CONFIG_FILE paths use recipies/ (typo) while existing B300 FP4 configs use recipes/. If the srt-slurm repo directory is actually named recipes/, all benchmark runs will fail to locate their config files. Please verify the actual directory name.

No other blocking issues found.

@jthomson04 @kedarpotdar-nv is it fine to stop this test and merge it?

seems like 2 1k1k and 4 8k1k of it has passed already? any concern that we need to test the full sweep?

@pr-claude review plz

Claude finished @functionstackx's task in 1m 40s —— View job

PR Review: [NV] DSR1 FP8 B300 Dynamo TRT

• Gather context and read changed files
• Validate master config + perf-changelog
• Validate container image accessibility
• Validate enroot import in launch script
• Validate model prefixes
• Check for bugs and logic errors
• Post review summary

Summary

LGTM - no blocking issues found.

The previous review's issues have both been fixed:

• recipies/ → recipes/ typo corrected in commit 28151c8
• Model prefix fixed in commit 489fbfb

All validations pass:

• perf-changelog.yaml updated alongside nvidia-master.yaml ✓
• Image nvcr.io/nvidia/ai-dynamo/tensorrtllm-runtime:0.8.1.post1 is publicly accessible on NGC ✓
• enroot import docker://$IMAGE pattern present in launch script ✓
• Model prefix dsr1 is valid across all 27 configs ✓
• All CONFIG_FILE paths use recipes/ (no typo remnants) ✓
• No duplicate CONFIG_FILE paths ✓
• Launch script FP8 handling (SRT_SLURM_MODEL_PREFIX, model path, served model name) mirrors the existing GB200 pattern exactly ✓