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": 76886.06691899558,
-      "required_pages_per_second_upper_estimate": 0.005319156509710125
+      "required_token_positions_per_second": 76802.21288301589,
+      "required_pages_per_second_upper_estimate": 0.00531335529293288
     },
     {
       "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": 9610.758364874448,
-      "required_pages_per_second_upper_estimate": 0.004023823831916432
+      "required_token_positions_per_second": 9600.276610376986,
+      "required_pages_per_second_upper_estimate": 0.0040194353401922685
     },
     {
       "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": 19221.516729748895,
-      "required_pages_per_second_upper_estimate": 0.004003335029322614
+      "required_token_positions_per_second": 19200.553220753973,
+      "required_pages_per_second_upper_estimate": 0.003998968883243881
     },
     {
       "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": 28832.275094623343,
-      "required_pages_per_second_upper_estimate": 0.0017284508675876917
+      "required_token_positions_per_second": 28800.829831130963,
+      "required_pages_per_second_upper_estimate": 0.0017265657720554606
     },
     {
       "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": 28832.275094623343,
-      "required_pages_per_second_upper_estimate": 0.0022415825688315506
+      "required_token_positions_per_second": 28800.829831130963,
+      "required_pages_per_second_upper_estimate": 0.002239137837908115
     },
     {
       "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": 28832.275094623343,
-      "required_pages_per_second_upper_estimate": 0.36982946722878546
+      "required_token_positions_per_second": 28800.829831130963,
+      "required_pages_per_second_upper_estimate": 0.3694261211519986
     }
   ],
   "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": 0.1243443915547573,
-    "producer_p95_seconds": 0.2668740259978222,
-    "gpu_update_seconds": 0.958925367813094,
-    "producer_batches_per_second": 8.042180169900401,
-    "gpu_batches_per_second": 1.0428340239664116,
-    "producer_headroom": 7.711850577440913,
+    "producer_batch_seconds": 0.12223035409377303,
+    "producer_p95_seconds": 0.2720489700004691,
+    "gpu_update_seconds": 0.9599723397592919,
+    "producer_batches_per_second": 8.18127385307922,
+    "gpu_batches_per_second": 1.041696680813475,
+    "producer_headroom": 7.853796602951976,
     "sustainable": true,
-    "buffer_seconds": 60.41229817222492,
+    "buffer_seconds": 60.47825740483539,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 318817.51138996455,
-    "upload_bytes_per_second": 3949280.9851866667,
+    "generation_bytes_per_second": 311256.93683523545,
+    "upload_bytes_per_second": 3924236.4582928116,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.41081625246241,
+    "estimated_full_shards_per_hour": 66.78849295418546,
     "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": 1.65895906500009,
-      "bytes": 12493410,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 0.8486537320022762,
       "bytes": 1121895,
@@ -562,6 +556,12 @@
       "bytes": 16198178,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0830437309996341,
+      "bytes": 7432031,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -283,3 +283,4 @@
 {"step": 553300, "loss": 0.6275357604026794, "lr": 0.0003, "grad_norm": 0.17894805029477764, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80985417451, "tokens_per_second": 149258.7486075883, "data_wait_s": 1.2778997188434005e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 553400, "loss": 0.5868434906005859, "lr": 0.0003, "grad_norm": 0.18740991507098057, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81000037393, "tokens_per_second": 151778.89971913135, "data_wait_s": 1.3491000572685152e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 553500, "loss": 0.5721019506454468, "lr": 0.0003, "grad_norm": 0.13858546248677567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81014783354, "tokens_per_second": 153291.6785604879, "data_wait_s": 1.3586002751253545e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 553600, "loss": 0.5862679481506348, "lr": 0.0003, "grad_norm": 0.17522218260013342, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 81029518079, "tokens_per_second": 152876.62346661417, "data_wait_s": 1.2905002222396433e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b3e450accfcd0235590e68030829bea90db3549c09378c67b5ea357a359495a5
-size 607710
+oid sha256:1e267d45b6d619f2c746ad77c651f8c316c7c382b8ea44a785185c41f9013981
+size 609828