Run Claude Code with any foundation model on Amazon Bedrock (43 models) or a self-hosted model on EC2. Includes a HumanEval benchmark.
aws-samples/sample-claude-code-multi-model
Skill Claude CodeCodex
Implement code from an LLD spec using the current model. Reads the best LLD for a given problem, implements the code changes against the target repo, produces a unified diff, and captures detailed metrics (tokens, cache hits, wall-clock time, tool calls). Results go under…
aws-samples/sample-claude-code-multi-model
Skill Claude CodeCodex
Summarize a /swe benchmark run for a given repo/problem/model triple. Reports artifact completion status (github-issue.md, lld.md, review.md, testing.md), error signals captured during the run, and a token-and-themes summary derived from the matching session JSONL files under /.claude/projects/. Output is a single…
aws-samples/sample-claude-code-multi-model
Skill Claude CodeCodex
End-to-end Software Engineering skill that benchmarks how well a given LLM can take a problem from idea to a complete design package. Creates structured documentation under benchmarks/swe-benchmark-data/{repo-name}/{problem-name}/{model-name}/ with a GitHub issue spec, low-level design (LLD), expert review, and…
aws-samples/sample-claude-code-multi-model
Skill Claude CodeCodex
Stand up a vLLM inference server for an open-weight coding model on a multi-GPU EC2 node (reference: g6e.12xlarge, 4xL40S). Drives the full flow end to end — verify the GPU node, install vLLM and its OS/Python dependencies (including the two Deep Learning AMI-specific fixes), serve a model with tensor parallelism and…