@aws/ml-container-creator 1.0.2 → 1.0.4

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (49) hide show
  1. package/README.md +1 -1
  2. package/bin/cli.js +1 -1
  3. package/config/tune-catalog.json +303 -1
  4. package/infra/ci-harness/lib/ci-harness-stack.ts +43 -0
  5. package/package.json +3 -2
  6. package/servers/base-image-picker/index.js +65 -18
  7. package/servers/instance-sizer/index.js +32 -0
  8. package/servers/lib/catalogs/fleet-drivers.json +38 -0
  9. package/servers/lib/catalogs/model-arch-support.json +51 -0
  10. package/servers/lib/catalogs/model-servers.json +2842 -1516
  11. package/servers/lib/schemas/image-catalog.schema.json +12 -0
  12. package/src/app.js +6 -4
  13. package/src/lib/bootstrap-command-handler.js +12 -2
  14. package/src/lib/bootstrap-profile-manager.js +16 -0
  15. package/src/lib/cross-cutting-checker.js +6 -1
  16. package/src/lib/generated/cli-options.js +1 -1
  17. package/src/lib/generated/parameter-matrix.js +1 -1
  18. package/src/lib/generated/validation-rules.js +1 -1
  19. package/src/lib/mcp-query-runner.js +110 -3
  20. package/src/lib/prompt-runner.js +66 -22
  21. package/src/lib/template-variable-resolver.js +8 -0
  22. package/src/lib/train-config-builder.js +339 -0
  23. package/templates/do/.benchmark_writer.py +3 -0
  24. package/templates/do/.eval_helper.py +409 -0
  25. package/templates/do/.register_helper.py +185 -11
  26. package/templates/do/.train_build_request.py +102 -113
  27. package/templates/do/.train_helper.py +433 -0
  28. package/templates/do/__pycache__/.register_helper.cpython-312.pyc +0 -0
  29. package/templates/do/adapter +157 -0
  30. package/templates/do/benchmark +60 -3
  31. package/templates/do/deploy.d/managed-inference.ejs +83 -0
  32. package/templates/do/evaluate +272 -0
  33. package/templates/do/lib/resolve-instance.sh +155 -0
  34. package/templates/do/register +5 -0
  35. package/templates/do/test +1 -0
  36. package/templates/do/train +879 -126
  37. package/templates/do/training/config.yaml +83 -11
  38. package/templates/do/training/dpo/accelerate_config.yaml +24 -0
  39. package/templates/do/training/dpo/defaults.yaml +26 -0
  40. package/templates/do/training/dpo/prompts.json +8 -0
  41. package/templates/do/training/dpo/train.py +363 -0
  42. package/templates/do/training/sft/accelerate_config.yaml +22 -0
  43. package/templates/do/training/sft/defaults.yaml +18 -0
  44. package/templates/do/training/sft/prompts.json +7 -0
  45. package/templates/do/training/sft/train.py +310 -0
  46. package/templates/do/tune +11 -2
  47. package/templates/do/.train_poll_parser.py +0 -135
  48. package/templates/do/.train_status_parser.py +0 -187
  49. /package/templates/do/training/{train.py → custom/train.py} +0 -0
package/servers/lib/catalogs/model-arch-support.json ADDED
@@ -0,0 +1,51 @@
1
+ {
2
+ "_comment": "Model architecture → minimum framework version mapping. Used by driver-aware filtering to exclude framework versions that don't support a given model architecture. Source: vLLM/SGLang release notes and supported_models docs.",
3
+ "_last_updated": "2026-06-29",
4
+ "vllm": {
5
+ "LlamaForCausalLM": "v0.4.0",
6
+ "Llama4ForCausalLM": "v0.22.0",
7
+ "MistralForCausalLM": "v0.4.0",
8
+ "MixtralForCausalLM": "v0.4.0",
9
+ "Qwen2ForCausalLM": "v0.6.0",
10
+ "Qwen2MoeForCausalLM": "v0.6.0",
11
+ "Qwen3ForCausalLM": "v0.20.0",
12
+ "Qwen3MoeForCausalLM": "v0.20.0",
13
+ "DeepseekV2ForCausalLM": "v0.16.0",
14
+ "DeepseekV3ForCausalLM": "v0.19.0",
15
+ "Gemma2ForCausalLM": "v0.8.0",
16
+ "Gemma3ForCausalLM": "v0.20.0",
17
+ "Gemma4ForCausalLM": "v0.23.0",
18
+ "GptOssForCausalLM": "v0.22.0",
19
+ "NemotronForCausalLM": "v0.17.0",
20
+ "Phi3ForCausalLM": "v0.6.0",
21
+ "PhiMoEForCausalLM": "v0.16.0",
22
+ "GraniteForCausalLM": "v0.17.0",
23
+ "GraniteMoeForCausalLM": "v0.19.0",
24
+ "CohereForCausalLM": "v0.16.0",
25
+ "Cohere2ForCausalLM": "v0.19.0",
26
+ "FalconForCausalLM": "v0.4.0",
27
+ "StarCoder2ForCausalLM": "v0.6.0",
28
+ "InternLM2ForCausalLM": "v0.6.0",
29
+ "OlmoForCausalLM": "v0.16.0",
30
+ "Olmo2ForCausalLM": "v0.19.0"
31
+ },
32
+ "sglang": {
33
+ "LlamaForCausalLM": "v0.3.0",
34
+ "MistralForCausalLM": "v0.3.0",
35
+ "Qwen2ForCausalLM": "v0.4.0",
36
+ "Qwen3ForCausalLM": "v0.5.0",
37
+ "Qwen3MoeForCausalLM": "v0.5.0",
38
+ "DeepseekV2ForCausalLM": "v0.4.0",
39
+ "DeepseekV3ForCausalLM": "v0.5.0",
40
+ "Gemma2ForCausalLM": "v0.4.0",
41
+ "Gemma3ForCausalLM": "v0.5.0",
42
+ "Phi3ForCausalLM": "v0.4.0",
43
+ "InternLM2ForCausalLM": "v0.4.0"
44
+ },
45
+ "lmi": {
46
+ "_comment": "DJL LMI uses HuggingFace transformers directly — architecture support is determined by the bundled transformers version, not the LMI version itself. No version-based filtering needed for LMI.",
47
+ "LlamaForCausalLM": "v0.25.0",
48
+ "Qwen2ForCausalLM": "v0.27.0",
49
+ "DeepseekV2ForCausalLM": "v0.28.0"
50
+ }
51
+ }