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": 76741.193642006,
-      "required_pages_per_second_upper_estimate": 0.0053091338402556695
+      "required_token_positions_per_second": 77170.14605582153,
+      "required_pages_per_second_upper_estimate": 0.005338809763550155
     },
     {
       "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": 9592.64920525075,
-      "required_pages_per_second_upper_estimate": 0.004016241904944244
+      "required_token_positions_per_second": 9646.268256977692,
+      "required_pages_per_second_upper_estimate": 0.0040386910822091905
     },
     {
       "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": 19185.2984105015,
-      "required_pages_per_second_upper_estimate": 0.003995791708564691
+      "required_token_positions_per_second": 19292.536513955383,
+      "required_pages_per_second_upper_estimate": 0.004018126577455167
     },
     {
       "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": 28777.947615752248,
-      "required_pages_per_second_upper_estimate": 0.0017251940181826266
+      "required_token_positions_per_second": 28938.804770933075,
+      "required_pages_per_second_upper_estimate": 0.0017348371590210627
     },
     {
       "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": 28777.947615752248,
-      "required_pages_per_second_upper_estimate": 0.0022373588463106484
+      "required_token_positions_per_second": 28938.804770933075,
+      "required_pages_per_second_upper_estimate": 0.0022498647825901067
     },
     {
       "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": 28777.947615752248,
-      "required_pages_per_second_upper_estimate": 0.3691326126621291
+      "required_token_positions_per_second": 28938.804770933075,
+      "required_pages_per_second_upper_estimate": 0.37119591553383197
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.17091287049221648,
-    "producer_p95_seconds": 0.1989745259998017,
-    "gpu_update_seconds": 0.9607356427623162,
-    "producer_batches_per_second": 5.850934438817121,
-    "gpu_batches_per_second": 1.0408690543891872,
-    "producer_headroom": 5.6212012588371385,
+    "producer_batch_seconds": 0.07858506867191295,
+    "producer_p95_seconds": 0.21533125899986771,
+    "gpu_update_seconds": 0.9553953668387302,
+    "producer_batches_per_second": 12.725063639950848,
+    "gpu_batches_per_second": 1.0466870938560864,
+    "producer_headroom": 12.157466844337028,
     "sustainable": true,
-    "buffer_seconds": 60.52634549402592,
+    "buffer_seconds": 60.18990811084,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 377921.38817357447,
-    "upload_bytes_per_second": 4792667.769900379,
+    "generation_bytes_per_second": 430143.458836989,
+    "upload_bytes_per_second": 4757232.520941385,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 81.0931323423903,
+    "estimated_full_shards_per_hour": 92.29877303917172,
     "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": 1.6006398999998055,
-      "bytes": 9959535,
-      "pages": 29,
-      "shards": 3
-    },
     {
       "seconds": 1.2196721769996657,
       "bytes": 7030636,
@@ -562,6 +556,12 @@
       "bytes": 7314599,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 1.2749757900000986,
+      "bytes": 5237846,
+      "pages": 31,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -126,3 +126,4 @@
 {"step": 855100, "loss": 0.5495004057884216, "lr": 3e-05, "grad_norm": 0.29392040575456885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125162672109, "tokens_per_second": 144075.6618093551, "data_wait_s": 1.6910998965613544e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 855200, "loss": 0.5429491996765137, "lr": 3e-05, "grad_norm": 0.2991947098934205, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125177218265, "tokens_per_second": 150919.99109127533, "data_wait_s": 1.471599898650311e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 855300, "loss": 0.5383842587471008, "lr": 3e-05, "grad_norm": 0.301649517380114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125191707924, "tokens_per_second": 150679.14492612047, "data_wait_s": 1.392699959978927e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 855400, "loss": 0.5500805974006653, "lr": 3e-05, "grad_norm": 0.3322299964541384, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125206396372, "tokens_per_second": 152831.21924984825, "data_wait_s": 1.70599996636156e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e50db9dc6b18c77f943fb87a633640e7ac5c7fdfe3a46087d518da420bc1547a
-size 273618
+oid sha256:6ccd2eac51dc1fcef88338203c2af17f95e8533414bc8566f0a6eac43478230d
+size 275736