8000 Repair verl OPD semantic contracts by DaoyuanLi2816 · Pull Request #75 · DaoyuanLi2816/mini-verl · GitHub
[go: up one dir, main page]

Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
136 changes: 92 additions & 44 deletions docs/generated/verl-opd-v0.8-compatibility.json
F8DB
Original file line number Diff line number Diff line change
@@ -1,13 +1,13 @@
{
"classification_counts": {
"exact": 23,
"informational_only": 12,
"locally_reinterpreted": 22,
"semantically_conformant": 15
"exact": 34,
"informational_only": 9,
"locally_reinterpreted": 20,
"semantically_conformant": 14
},
"compiled_digest": "d9458fba483548bedbff9b112d8e192e48055baec24a389f151323417d3072ea",
"compiled_digest": "860b25e12c88bc772c9003008aafece2e4a71337a75373441f940ddde65bd008",
"executable": true,
"field_count": 72,
"field_count": 77,
"fields": [
{
"classification": "semantically_conformant",
Expand Down Expand Up @@ -60,14 +60,14 @@
"user_confirmation_required": false
},
{
"classification": "locally_reinterpreted",
"classification": "informational_only",
"executable": true,
"local_target": "batching.update_trajectory_batch_size",
"reason": "used as a logical update batch, not a PPO mini-batch",
"semantic_risk": "high",
"local_target": null,
"reason": "direct one-step GKD has no PPO mini-batch; data.train_batch_size is the logical group",
"semantic_risk": "none",
"source_value": 8,
"upstream_field": "actor_rollout_ref.actor.ppo_mini_batch_size",
"user_confirmation_required": true
"user_confirmation_required": false
},
{
"classification": "locally_reinterpreted",
Expand Down Expand Up @@ -100,10 +100,10 @@
"user_confirmation_required": false
},
{
"classification": "exact",
"classification": "informational_only",
"executable": true,
"local_target": "student.adapter.path",
"reason": "same optional adapter artifact identity",
"reason": "null records that no existing adapter input is configured",
"semantic_risk": "none",
"source_value": null,
"upstream_field": "actor_rollout_ref.model.lora_adapter_path",
Expand Down Expand Up @@ -253,10 +253,10 @@
"user_confirmation_required": false
},
{
"classification": "semantically_conformant",
"classification": "informational_only",
"executable": true,
"local_target": "source.filter_overlong_prompts",
"reason": "same filtering intent",
"local_target": null,
"reason": "false is recorded; explicit truncation controls local overlength handling",
"semantic_risk": "none",
"source_value": false,
"upstream_field": "data.filter_overlong_prompts",
Expand Down Expand Up @@ -557,10 +557,40 @@
"user_confirmation_required": true
},
{
"classification": "informational_only",
"classification": "exact",
"executable": true,
"local_target": null,
"reason": "miniVERL-only physical batch",
"local_target": "student.attn_implementation",
"reason": "miniVERL-only explicit actor attention implementation",
"semantic_risk": "none",
"source_value": "sdpa",
"upstream_field": "miniverl.actor_runtime.attn_implementation",
"user_confirmation_required": false
},
{
"classification": "exact",
"executable": true,
"local_target": "student.dtype",
"reason": "miniVERL-only explicit actor runtime dtype",
"semantic_risk": "none",
"source_value": "auto",
"upstream_field": "miniverl.actor_runtime.dtype",
"user_confirmation_required": false
},
{
"classification": "exact",
"executable": true,
"local_target": "student.quantization",
"reason": "miniVERL-only explicit actor quantization",
"semantic_risk": "none",
"source_value": "nf4",
"upstream_field": "miniverl.actor_runtime.quantization",
"user_confirmation_required": false
},
{
"classification": "exact",
"executable": true,
"local_target": "batching.rollout_batch_size",
"reason": "miniVERL-only physical rollout batch",
"semantic_risk": "none",
"source_value": 2,
"upstream_field": "miniverl.batching.rollout_batch_size",
Expand All @@ -570,54 +600,54 @@
"classification": "informational_only",
"executable": true,
"local_target": null,
"reason": "miniVERL-only physical batch",
"reason": "reserved for a future batched teacher scorer; current scoring is trajectory-serial",
"semantic_risk": "none",
"source_value": 2,
"upstream_field": "miniverl.batching.teacher_score_batch_size",
"user_confirmation_required": false
},
{
"classification": "informational_only",
"classification": "exact",
"executable": true,
"local_target": null,
"reason": "miniVERL-only physical batch",
"local_target": "batching.update_trajectory_batch_size",
"reason": "miniVERL-only actor-update batch",
"semantic_risk": "none",
"source_value": 2,
"upstream_field": "miniverl.batching.update_trajectory_batch_size",
"user_confirmation_required": false
},
{
"classification": "informational_only",
"classification": "exact",
"executable": true,
"local_target": null,
"local_target": "memory.headroom_gib",
"reason": "miniVERL-only planner headroom",
"semantic_risk": "none",
"source_value": 1.5,
"upstream_field": "miniverl.memory.headroom_gib",
"user_confirmation_required": false
},
{
"classification": "informational_only",
"classification": "exact",
"executable": true,
"local_target": null,
"local_target": "memory.vram_limit_gib",
"reason": "miniVERL-only planner limit",
"semantic_risk": "none",
"source_value": 16,
"upstream_field": "miniverl.memory.vram_limit_gib",
"user_confirmation_required": false
},
{
"classification": "informational_only",
"classification": "exact",
"executable": true,
"local_target": null,
"reason": "miniVERL-only local extension",
"local_target": "placement.requested",
"reason": "miniVERL-only requested local placement",
"semantic_risk": "none",
"source_value": "auto",
"upstream_field": "miniverl.runtime.mode",
"user_confirmation_required": false
},
{
"classification": "informational_only",
"classification": "exact",
"executable": true,
"local_target": "student.revision",
"reason": "miniVERL-only immutable Hub revision",
Expand All @@ -629,8 +659,8 @@
{
"classification": "informational_only",
"executable": true,
"local_target": null,
"reason": "miniVERL-only teacher adapter extension",
"local_target": "teacher.adapter.path",
"reason": "null records that no existing adapter input is configured",
"semantic_risk": "none",
"source_value": null,
"upstream_field": "miniverl.teacher_adapter.path",
Expand All @@ -639,15 +669,15 @@
{
"classification": "informational_only",
"executable": true,
"local_target": null,
"reason": "miniVERL-only teacher adapter extension",
"local_target": "teacher.adapter.revision",
"reason": "null records that no existing adapter input is configured",
"semantic_risk": "none",
"source_value": null,
"upstream_field": "miniverl.teacher_adapter.revision",
"user_confirmation_required": false
},
{
"classification": "informational_only",
"classification": "exact",
"executable": true,
"local_target": "teacher.revision",
"reason": "miniVERL-only immutable Hub revision",
Expand All @@ -656,6 +686,26 @@
"upstream_field": "miniverl.teacher_revision",
"user_confirmation_required": false
},
{
"classification": "exact",
"executable": true,
"local_target": "teacher.attn_implementation",
"reason": "miniVERL-only explicit teacher attention implementation",
"semantic_risk": "none",
"source_value": "sdpa",
"upstream_field": "miniverl.teacher_runtime.attn_implementation",
"user_confirmation_required": false
},
{
"classification": "exact",
"executable": true,
"local_target": "teacher.quantization",
"reason": "miniVERL-only explicit teacher quantization",
"semantic_risk": "none",
"source_value": "nf4",
"upstream_field": "miniverl.teacher_runtime.quantization",
"user_confirmation_required": false
},
{
"classification": "exact",
"executable": true,
Expand Down Expand Up @@ -717,14 +767,14 @@
"user_confirmation_required": true
},
{
"classification": "locally_reinterpreted",
"classification": "informational_only",
"executable": true,
"local_target": "schedule.dataset_passes",
"reason": "local bounded dataset passes",
"semantic_risk": "medium",
"reason": "recorded but superseded by the explicit total_training_steps cap",
"semantic_risk": "none",
"source_value": 1,
"upstream_field": "trainer.total_epochs",
"user_confirmation_required": true
"user_confirmation_required": false
},
{
"classification": "semantically_conformant",
Expand All @@ -741,7 +791,6 @@
"reinterpretation_acceptance": {
"accepted": false,
"required_fields": [
"actor_rollout_ref.actor.ppo_mini_batch_size",
"actor_rollout_ref.rollout.gpu_memory_utilization",
"actor_rollout_ref.rollout.name",
"actor_rollout_ref.rollout.tensor_model_parallel_size",
Expand All @@ -756,15 +805,14 @@
"trainer.n_gpus_per_node",
"trainer.nnodes",
"trainer.save_freq",
"trainer.test_freq",
"trainer.total_epochs"
"trainer.test_freq"
],
"source": "not_accepted"
},
"schema_version": 2,
"scope": "config conformance for one documented local pure-OPD profile; not full verl compatibility or distributed execution",
"source_fixture": "examples/verl-opd-v0.8-single-gpu.yaml",
"source_sha256": "085805d904136ae271db7629b4810340f54d6aba7c2ed08e7a88d8ac01511b4f",
"source_sha256": "ed8dcbc2e7edefb91beea3f74234a27a8f724411649880b897c9adaffdeb4182",
"upstream": {
"commit": "7aed6b230776f963fa09509c10d9c3a767d1102c",
"repository": "https://github.com/verl-project/verl",
Expand Down
Loading
0