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": 75598.86968125336,
-      "required_pages_per_second_upper_estimate": 0.005230105218094038
+      "required_token_positions_per_second": 75827.00156316662,
+      "required_pages_per_second_upper_estimate": 0.005245887911023692
     },
     {
       "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": 9449.85871015667,
-      "required_pages_per_second_upper_estimate": 0.003956458506452963
+      "required_token_positions_per_second": 9478.375195395827,
+      "required_pages_per_second_upper_estimate": 0.003968397763330671
     },
     {
       "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": 18899.71742031334,
-      "required_pages_per_second_upper_estimate": 0.0039363127195856645
+      "required_token_positions_per_second": 18956.750390791654,
+      "required_pages_per_second_upper_estimate": 0.003948191183275722
     },
     {
       "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": 28349.576130470006,
-      "required_pages_per_second_upper_estimate": 0.0016995138017253414
+      "required_token_positions_per_second": 28435.125586187478,
+      "required_pages_per_second_upper_estimate": 0.0017046423609691484
     },
     {
       "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": 28349.576130470006,
-      "required_pages_per_second_upper_estimate": 0.002204054847536989
+      "required_token_positions_per_second": 28435.125586187478,
+      "required_pages_per_second_upper_estimate": 0.002210705941426734
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28349.576130470006,
-      "required_pages_per_second_upper_estimate": 0.36363792319839416
+      "required_token_positions_per_second": 28435.125586187478,
+      "required_pages_per_second_upper_estimate": 0.36473525976048893
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.23268139603123217,
-    "producer_p95_seconds": 0.18061906499951874,
-    "gpu_update_seconds": 0.9752526765394578,
-    "producer_batches_per_second": 4.297722194626049,
-    "gpu_batches_per_second": 1.0253752940708192,
-    "producer_headroom": 4.191365073332087,
+    "producer_batch_seconds": 0.07508604669531849,
+    "producer_p95_seconds": 0.1859565240001757,
+    "gpu_update_seconds": 0.9723185472207012,
+    "producer_batches_per_second": 13.318053673244574,
+    "gpu_batches_per_second": 1.0284695307504153,
+    "producer_headroom": 12.949390599376487,
     "sustainable": true,
-    "buffer_seconds": 60.46566594544638,
+    "buffer_seconds": 60.28374992768347,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 438343.8641476351,
-    "upload_bytes_per_second": 5433148.26919615,
+    "generation_bytes_per_second": 456986.4307449575,
+    "upload_bytes_per_second": 5444060.92887007,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.05838912317076,
+    "estimated_full_shards_per_hour": 98.05864993821663,
     "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": [
@@ -508,6 +508,12 @@
       "bytes": 7441828,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.1509858810004516,
+      "bytes": 6964137,
+      "pages": 33,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -64,3 +64,4 @@
 {"step": 610800, "loss": 0.5893732309341431, "lr": 0.0001, "grad_norm": 0.15391375382795708, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89403323012, "tokens_per_second": 150141.51985255766, "data_wait_s": 1.0426999324408825e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 610900, "loss": 0.5411556363105774, "lr": 0.0001, "grad_norm": 0.1468534848009401, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89417853373, "tokens_per_second": 150565.84804299244, "data_wait_s": 1.4136000572761986e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 611000, "loss": 0.5577870011329651, "lr": 0.0001, "grad_norm": 0.15734135282543868, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89432608088, "tokens_per_second": 153349.19443355923, "data_wait_s": 1.5098999938345514e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 611100, "loss": 0.5393580198287964, "lr": 0.0001, "grad_norm": 0.14541632655303988, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89447318702, "tokens_per_second": 152162.13442943935, "data_wait_s": 1.2389000403345563e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0bf853712f1be7cbbf0016c363e02995357c4fb9b5d47ac98755a72518c4dafc
-size 141519
+oid sha256:739fcdbd558ef114da0a2f99a611566b8b055d188eba0fe5f3f4ab7b08d7314e
+size 143637