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": 76852.25315825094,
-      "required_pages_per_second_upper_estimate": 0.005316817195283076
+      "required_token_positions_per_second": 77666.70148179415,
+      "required_pages_per_second_upper_estimate": 0.005373162620086267
     },
     {
       "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": 9606.531644781368,
-      "required_pages_per_second_upper_estimate": 0.004022054192477411
+      "required_token_positions_per_second": 9708.337685224269,
+      "required_pages_per_second_upper_estimate": 0.00406467825565898
     },
     {
       "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": 19213.063289562735,
-      "required_pages_per_second_upper_estimate": 0.004001574400664046
+      "required_token_positions_per_second": 19416.675370448538,
+      "required_pages_per_second_upper_estimate": 0.004043981427500895
     },
     {
       "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": 28819.594934344103,
-      "required_pages_per_second_upper_estimate": 0.0017276907113404347
+      "required_token_positions_per_second": 29125.01305567281,
+      "required_pages_per_second_upper_estimate": 0.0017460000613676219
     },
     {
       "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": 28819.594934344103,
-      "required_pages_per_second_upper_estimate": 0.0022405967421439705
+      "required_token_positions_per_second": 29125.01305567281,
+      "required_pages_per_second_upper_estimate": 0.002264341657684936
     },
     {
       "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": 28819.594934344103,
-      "required_pages_per_second_upper_estimate": 0.3696668197476187
+      "required_token_positions_per_second": 29125.01305567281,
+      "required_pages_per_second_upper_estimate": 0.37358439547559436
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08620114205484697,
-    "producer_p95_seconds": 0.30581583800085355,
-    "gpu_update_seconds": 0.9593472796195369,
-    "producer_batches_per_second": 11.600774376791119,
-    "gpu_batches_per_second": 1.0423753954840893,
-    "producer_headroom": 11.129171339854588,
+    "producer_batch_seconds": 0.3588905510005702,
+    "producer_p95_seconds": 0.30710258500039345,
+    "gpu_update_seconds": 0.9492871281173512,
+    "producer_batches_per_second": 2.786364804567984,
+    "gpu_batches_per_second": 1.0534220578585363,
+    "producer_headroom": 2.6450602432156063,
     "sustainable": true,
-    "buffer_seconds": 60.43887861603083,
+    "buffer_seconds": 60.754376199510475,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 366692.5772180197,
-    "upload_bytes_per_second": 5010787.708689514,
+    "generation_bytes_per_second": 346211.5581092768,
+    "upload_bytes_per_second": 5007355.727547088,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.68369090466922,
+    "estimated_full_shards_per_hour": 74.28894097765664,
     "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": 0.9105693140008952,
-      "bytes": 1064335,
-      "pages": 31,
-      "shards": 1
-    },
     {
       "seconds": 1.423822492997715,
       "bytes": 7690011,
@@ -562,6 +556,12 @@
       "bytes": 7581932,
       "pages": 31,
       "shards": 3
+    },
+    {
+      "seconds": 0.9802855540037854,
+      "bytes": 1121294,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -373,3 +373,4 @@
 {"step": 922300, "loss": 0.49285024404525757, "lr": 3e-05, "grad_norm": 0.33830968570608894, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134998494351, "tokens_per_second": 154051.0488153253, "data_wait_s": 1.481799699831754e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 922400, "loss": 0.5174910426139832, "lr": 3e-05, "grad_norm": 0.3362519461787307, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135013189439, "tokens_per_second": 153450.3945819103, "data_wait_s": 1.3343000318855047e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 922500, "loss": 0.5190991759300232, "lr": 3e-05, "grad_norm": 0.3377101964376676, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135027664212, "tokens_per_second": 151365.4228573738, "data_wait_s": 1.276100374525413e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 922600, "loss": 0.5873844027519226, "lr": 3e-05, "grad_norm": 0.3420149008279359, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135042279656, "tokens_per_second": 151919.74790768157, "data_wait_s": 1.654799416428432e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b04b920247c609072a12a93ac7d874f56884f621301d9ae26134ef5cc7c98665
-size 799896
+oid sha256:68ea94ed62e89a1bc601155b12d1e7913cf2809f51e86ed157949e6d8b9e980d
+size 802014