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": 76879.25394647518,
-      "required_pages_per_second_upper_estimate": 0.005318685172462889
+      "required_token_positions_per_second": 77399.60978839574,
+      "required_pages_per_second_upper_estimate": 0.005354684596998853
     },
     {
       "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": 9609.906743309397,
-      "required_pages_per_second_upper_estimate": 0.0040234672757510255
+      "required_token_positions_per_second": 9674.951223549468,
+      "required_pages_per_second_upper_estimate": 0.00405070004134434
     },
     {
       "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": 19219.813486618794,
-      "required_pages_per_second_upper_estimate": 0.004002980288696156
+      "required_token_positions_per_second": 19349.902447098935,
+      "required_pages_per_second_upper_estimate": 0.004030074388487579
     },
     {
       "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": 28829.72022992819,
-      "required_pages_per_second_upper_estimate": 0.0017282977073502697
+      "required_token_positions_per_second": 29024.8536706484,
+      "required_pages_per_second_upper_estimate": 0.0017399956591699344
     },
     {
       "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": 28829.72022992819,
-      "required_pages_per_second_upper_estimate": 0.0022413839393391646
+      "required_token_positions_per_second": 29024.8536706484,
+      "required_pages_per_second_upper_estimate": 0.0022565547060538634
     },
     {
       "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": 28829.72022992819,
-      "required_pages_per_second_upper_estimate": 0.369796696167676
+      "required_token_positions_per_second": 29024.8536706484,
+      "required_pages_per_second_upper_estimate": 0.3722996584272701
     }
   ],
   "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": 0.08997315123446015,
-    "producer_p95_seconds": 0.24526039400006994,
-    "gpu_update_seconds": 0.9590103469439345,
-    "producer_batches_per_second": 11.114426762647335,
-    "gpu_batches_per_second": 1.0427416171125647,
-    "producer_headroom": 10.658850265729372,
+    "producer_batch_seconds": 0.15978182482811576,
+    "producer_p95_seconds": 0.25138754199724644,
+    "gpu_update_seconds": 0.9525629418748541,
+    "producer_batches_per_second": 6.258534104712744,
+    "gpu_batches_per_second": 1.0497993949163917,
+    "producer_headroom": 5.961647658609278,
     "sustainable": true,
-    "buffer_seconds": 60.417651857467874,
+    "buffer_seconds": 60.01146533811581,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318272.3164589689,
-    "upload_bytes_per_second": 5514562.881417241,
+    "generation_bytes_per_second": 346973.7765264379,
+    "upload_bytes_per_second": 5546989.657525898,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.29383011175919,
+    "estimated_full_shards_per_hour": 74.45249530644276,
     "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": 1.1689961140000378,
-      "bytes": 10341916,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 4.051943458998721,
       "bytes": 2586599,
@@ -562,6 +556,12 @@
       "bytes": 5745742,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.232796011998289,
+      "bytes": 13189628,
+      "pages": 37,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -259,3 +259,4 @@
 {"step": 583000, "loss": 0.553880512714386, "lr": 0.0003, "grad_norm": 0.16887774106607287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85334310790, "tokens_per_second": 151980.73036090282, "data_wait_s": 1.0558000212768093e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 583100, "loss": 0.5865918397903442, "lr": 0.0003, "grad_norm": 0.1487584477832386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85348969633, "tokens_per_second": 152443.44046201778, "data_wait_s": 1.1420001101214439e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 583200, "loss": 0.579066276550293, "lr": 0.0003, "grad_norm": 0.15166410431059976, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85363615908, "tokens_per_second": 153229.74095765813, "data_wait_s": 1.51149979501497e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 583300, "loss": 0.5727190375328064, "lr": 0.0003, "grad_norm": 0.17853147238948933, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 85378311432, "tokens_per_second": 152224.17760109832, "data_wait_s": 9.898998541757464e-06, "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:d0ef2e88d872accdee4cee56701caa68c2dce6658402e978e25b0bd9b1730a94
-size 556878
+oid sha256:93459dd4639e9ee42a60c110e5fc735fa998c9e9cb4211d3aa38e67764d99c74
+size 558996