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": 76169.50566937983,
-      "required_pages_per_second_upper_estimate": 0.0052695831398105394
+      "required_token_positions_per_second": 75984.30839086523,
+      "required_pages_per_second_upper_estimate": 0.00525677076236601
     },
     {
       "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": 9521.188208672478,
-      "required_pages_per_second_upper_estimate": 0.003986322677952223
+      "required_token_positions_per_second": 9498.038548858154,
+      "required_pages_per_second_upper_estimate": 0.003976630398807834
     },
     {
       "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": 19042.376417344956,
-      "required_pages_per_second_upper_estimate": 0.003966024826496603
+      "required_token_positions_per_second": 18996.07709771631,
+      "required_pages_per_second_upper_estimate": 0.003956381899213123
     },
     {
       "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": 28563.564626017433,
-      "required_pages_per_second_upper_estimate": 0.0017123420852918959
+      "required_token_positions_per_second": 28494.115646574464,
+      "required_pages_per_second_upper_estimate": 0.0017081787250167414
     },
     {
       "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": 28563.564626017433,
-      "required_pages_per_second_upper_estimate": 0.002220691511829882
+      "required_token_positions_per_second": 28494.115646574464,
+      "required_pages_per_second_upper_estimate": 0.0022152921591519683
     },
     {
       "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": 28563.564626017433,
-      "required_pages_per_second_upper_estimate": 0.36638273785633113
+      "required_token_positions_per_second": 28494.115646574464,
+      "required_pages_per_second_upper_estimate": 0.36549192091654115
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.09540528670322601,
-    "producer_p95_seconds": 0.22128988500116975,
-    "gpu_update_seconds": 0.967946415722095,
-    "producer_batches_per_second": 10.481599443337622,
-    "gpu_batches_per_second": 1.033115040003524,
-    "producer_headroom": 10.145626612213357,
+    "producer_batch_seconds": 0.08586985720310736,
+    "producer_p95_seconds": 0.22893109399956302,
+    "gpu_update_seconds": 0.9703056007398432,
+    "producer_batches_per_second": 11.645530021491792,
+    "gpu_batches_per_second": 1.0306031411521435,
+    "producer_headroom": 11.299723003437473,
     "sustainable": true,
-    "buffer_seconds": 60.01267777476989,
+    "buffer_seconds": 60.15894724587028,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 468299.689266289,
-    "upload_bytes_per_second": 5153462.617180967,
+    "generation_bytes_per_second": 476879.61052465654,
+    "upload_bytes_per_second": 5132874.784734549,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 100.48621185771468,
+    "estimated_full_shards_per_hour": 102.32726322941564,
     "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": 0.9075711420000516,
-      "bytes": 7428003,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 1.278834224000093,
       "bytes": 9834674,
@@ -562,6 +556,12 @@
       "bytes": 13342349,
       "pages": 36,
       "shards": 4
+    },
+    {
+      "seconds": 1.0075590680007736,
+      "bytes": 6216793,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -153,3 +153,4 @@
 {"step": 455500, "loss": 0.6034132242202759, "lr": 0.0003, "grad_norm": 0.1672609640580655, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66671161434, "tokens_per_second": 150856.0205691735, "data_wait_s": 1.2145999789936468e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 455600, "loss": 0.5722441077232361, "lr": 0.0003, "grad_norm": 0.13953617455733194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66685849006, "tokens_per_second": 152365.03168072176, "data_wait_s": 1.671499921940267e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 455700, "loss": 0.5648539662361145, "lr": 0.0003, "grad_norm": 0.18789220262180192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66700443276, "tokens_per_second": 151517.46830350376, "data_wait_s": 1.917999907163903e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 455800, "loss": 0.5343227386474609, "lr": 0.0003, "grad_norm": 0.15610936897271188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66715125888, "tokens_per_second": 151812.79188532886, "data_wait_s": 1.8380998881184496e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dbffc2c29e7b54905274bc973efb082dad5951e33be45e890781a3a5fa496fb3
-size 330804
+oid sha256:218afe33de43018fb9dbda43e7f67ade4495fbd9276cca5c96f4cb4f3dd1a673
+size 332922