Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77137.60248052294,
-      "required_pages_per_second_upper_estimate": 0.0053365583234995006
+      "required_token_positions_per_second": 76560.88933113698,
+      "required_pages_per_second_upper_estimate": 0.005296659969666111
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9642.200310065367,
-      "required_pages_per_second_upper_estimate": 0.004036987917785392
+      "required_token_positions_per_second": 9570.111166392122,
+      "required_pages_per_second_upper_estimate": 0.004006805698721907
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19284.400620130735,
-      "required_pages_per_second_upper_estimate": 0.004016432085329442
+      "required_token_positions_per_second": 19140.222332784244,
+      "required_pages_per_second_upper_estimate": 0.003986403550312294
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28926.6009301961,
-      "required_pages_per_second_upper_estimate": 0.0017341055574030667
+      "required_token_positions_per_second": 28710.333499176362,
+      "required_pages_per_second_upper_estimate": 0.0017211406551346804
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28926.6009301961,
-      "required_pages_per_second_upper_estimate": 0.0022489159876518294
+      "required_token_positions_per_second": 28710.333499176362,
+      "required_pages_per_second_upper_estimate": 0.0022321021461499417
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28926.6009301961,
-      "required_pages_per_second_upper_estimate": 0.3710393777683213
+      "required_token_positions_per_second": 28710.333499176362,
+      "required_pages_per_second_upper_estimate": 0.3682653313730758
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.4342880699142029,
-    "producer_p95_seconds": 0.35421874500025297,
-    "gpu_update_seconds": 0.9557984384932906,
-    "producer_batches_per_second": 2.302619089208594,
-    "gpu_batches_per_second": 1.0462456933664677,
-    "producer_headroom": 2.200839729910417,
+    "producer_batch_seconds": 0.4959902255625934,
+    "producer_p95_seconds": 0.34334457499790005,
+    "gpu_update_seconds": 0.9629982180734041,
+    "producer_batches_per_second": 2.016168763942307,
+    "gpu_batches_per_second": 1.0384235206588672,
+    "producer_headroom": 1.9415669270116993,
     "sustainable": true,
-    "buffer_seconds": 60.21530162507731,
+    "buffer_seconds": 60.66888773862446,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 291423.1978606891,
-    "upload_bytes_per_second": 5066393.846120054,
+    "generation_bytes_per_second": 269709.538346818,
+    "upload_bytes_per_second": 5075636.915534118,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.532634276061096,
+    "estimated_full_shards_per_hour": 57.87338841250806,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
-    {
-      "seconds": 1.1113367050020315,
-      "bytes": 10487773,
-      "pages": 35,
-      "shards": 3
-    },
     {
       "seconds": 1.1902512640008354,
       "bytes": 7677939,
@@ -562,6 +556,12 @@
       "bytes": 10956336,
       "pages": 35,
       "shards": 4
+    },
+    {
+      "seconds": 1.2077032870001858,
+      "bytes": 11770097,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -364,3 +364,4 @@
 {"step": 593500, "loss": 0.5857194662094116, "lr": 0.0003, "grad_norm": 0.1917992372402928, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86871096226, "tokens_per_second": 154434.70405764738, "data_wait_s": 1.2700002116616815e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 593600, "loss": 0.5902826189994812, "lr": 0.0003, "grad_norm": 0.1679454572026837, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86885610961, "tokens_per_second": 151177.0923937074, "data_wait_s": 1.3564000255428255e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 593700, "loss": 0.5640764236450195, "lr": 0.0003, "grad_norm": 0.1845438376884235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86900147008, "tokens_per_second": 150338.59316073087, "data_wait_s": 1.534100010758266e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 593800, "loss": 0.6211135387420654, "lr": 0.0003, "grad_norm": 0.16296979909993956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86914619690, "tokens_per_second": 138247.07420084206, "data_wait_s": 1.0039002518169582e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f553b015b64db4df67197b3b25e94c76a12a28dcef7856d30d2ff7c8ef0f345a
-size 780834
+oid sha256:e6a72cbf222d0375a28bd388d1afddb44943af50d740b86e6ad82778b94de44b
+size 782952