Dakarapu Varaha Surya Prakash
Community@progmastermind · visakhapatnam
Debugging, profiling, and optimization skills for the ATOM LLM inference engine on ROCm GPUs, plus VLSI agentic verification pipeline testing.
Agent Skills by Dakarapu Varaha Surya Prakash
Showing 7 vetted skills indexed across 2 GitHub repositories.
testing-vlsi-verify-agentic
Test the VLSI agentic verification pipeline end-to-end with deterministic and OpenAI-backed runs.
debug-agent-locate-kernel
Diagnoses faulty GPU kernels and silent hangs on ATOM platforms using rocm-debug-agent and rocgdb.
capture-trace
Capture and analyze PyTorch profiler and kineto traces from ATOM servers.
dump-bisect-debug
Compare intermediate tensor outputs between target and reference neural network implementations.
run-atom-workload
Automate ATOM workload execution for accuracy evaluation, performance benchmarking, and fault reproduction.
review-pr
Automate code review for ATOM PRs with risk assessments.
atom-patterns
Provide coding patterns and architecture guidance for optimizing the ATOM LLM inference engine.
Frequently Asked Questions About Dakarapu Varaha Surya Prakash
FAQPage SchemaWhat tasks can I perform with ProgMastermind's skills?▼
You can debug faulty GPU kernels and silent hangs with rocm-debug-agent and rocgdb, capture PyTorch profiler and kineto traces, bisect tensor outputs between neural network implementations, run ATOM workloads for accuracy and benchmarking, review ATOM PRs with risk assessments, and test the VLSI agentic verification pipeline.
Who should use these ATOM and VLSI skills?▼
These skills target GPU kernel engineers, LLM inference performance engineers, and hardware verification engineers working on the ATOM inference engine on ROCm platforms, plus verification engineers maintaining the VLSI agentic pipeline's orchestrator, agents, schemas, and simulator fallback.
How do I debug a hanging GPU kernel on ATOM hardware?▼
Use the debug-agent-locate-kernel skill, which drives rocm-debug-agent and rocgdb to locate faulty kernels and silent hangs. Complement it with capture-trace to collect PyTorch profiler and kineto traces from ATOM servers for deeper performance analysis.
What does the VLSI verification testing skill cover?▼
The testing-vlsi-verify-agentic skill tests the agentic verification pipeline end-to-end, covering orchestrator changes, agents, schemas, simulator fallback, OpenAI-backed demo audit, functional coverage extraction, the APB example flow, and the uart_lite official demo flow.
What prerequisites are needed to run these skills?▼
ATOM skills require access to ATOM servers with ROCm tooling (rocm-debug-agent, rocgdb) and PyTorch with kineto profiling. The VLSI skill requires the agentic verification pipeline repository including its orchestrator, schemas, simulator fallback, and demo flows.