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": 76594.96388485581,
-      "required_pages_per_second_upper_estimate": 0.005299017326356237
+      "required_token_positions_per_second": 77107.77088085262,
+      "required_pages_per_second_upper_estimate": 0.005334494504215461
     },
     {
       "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": 9574.370485606976,
-      "required_pages_per_second_upper_estimate": 0.004008588986732468
+      "required_token_positions_per_second": 9638.471360106578,
+      "required_pages_per_second_upper_estimate": 0.0040354266843069765
     },
     {
       "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": 19148.740971213952,
-      "required_pages_per_second_upper_estimate": 0.0039881777580456
+      "required_token_positions_per_second": 19276.942720213156,
+      "required_pages_per_second_upper_estimate": 0.004014878801454657
     },
     {
       "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": 28723.111456820927,
-      "required_pages_per_second_upper_estimate": 0.0017219066741846604
+      "required_token_positions_per_second": 28915.414080319733,
+      "required_pages_per_second_upper_estimate": 0.0017334349228343095
     },
     {
       "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": 28723.111456820927,
-      "required_pages_per_second_upper_estimate": 0.0022330955761525107
+      "required_token_positions_per_second": 28915.414080319733,
+      "required_pages_per_second_upper_estimate": 0.002248046259279694
     },
     {
       "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": 28723.111456820927,
-      "required_pages_per_second_upper_estimate": 0.3684292332938383
+      "required_token_positions_per_second": 28915.414080319733,
+      "required_pages_per_second_upper_estimate": 0.3708958848696109
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06757001773438986,
-    "producer_p95_seconds": 0.1506250460006413,
-    "gpu_update_seconds": 0.9625698121724338,
-    "producer_batches_per_second": 14.799463334920045,
-    "gpu_batches_per_second": 1.0388856863722846,
-    "producer_headroom": 14.245516642546807,
+    "producer_batch_seconds": 0.20650332182034248,
+    "producer_p95_seconds": 0.15453282500038767,
+    "gpu_update_seconds": 0.9561682195939102,
+    "producer_batches_per_second": 4.842537113616014,
+    "gpu_batches_per_second": 1.0458410764004533,
+    "producer_headroom": 4.630280090243656,
     "sustainable": true,
-    "buffer_seconds": 60.64189816686333,
+    "buffer_seconds": 60.23859783441634,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 494974.5960996609,
-    "upload_bytes_per_second": 2391530.751115987,
+    "generation_bytes_per_second": 462690.4746450342,
+    "upload_bytes_per_second": 2482454.252493539,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.21002590410586,
+    "estimated_full_shards_per_hour": 99.28260497582693,
     "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": [
@@ -250,6 +250,12 @@
       "bytes": 7819084,
       "pages": 35,
       "shards": 3
+    },
+    {
+      "seconds": 1.079906239000593,
+      "bytes": 5671187,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -30,3 +30,4 @@
 {"step": 528000, "loss": 0.5436892509460449, "lr": 0.0003, "grad_norm": 0.1983220608434485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77281614079, "tokens_per_second": 152006.00289369648, "data_wait_s": 1.8050000107905362e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 528100, "loss": 0.57123202085495, "lr": 0.0003, "grad_norm": 0.166388769357232, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77296205947, "tokens_per_second": 152000.73141821977, "data_wait_s": 8.165000508597586e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 528200, "loss": 0.5649438500404358, "lr": 0.0003, "grad_norm": 0.15943699597456476, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77310758756, "tokens_per_second": 151397.26698401864, "data_wait_s": 1.5206000171019696e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 528300, "loss": 0.5714080333709717, "lr": 0.0003, "grad_norm": 0.16946352881597554, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77325439916, "tokens_per_second": 153162.9703753375, "data_wait_s": 1.6467000023112632e-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 sha256:1eea85513c81fa0a8876fbef5160cfb496f4084534d9c7f59034a3ebe421e00e
-size 67941
+oid sha256:4a3d58db06b59c6c31b05e9498d895b615dd6ab202e69d510018bc8babb81fb8
+size 70059