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": 77672.00736694397,
-      "required_pages_per_second_upper_estimate": 0.005373529693532281
+      "required_token_positions_per_second": 76934.36213811544,
+      "required_pages_per_second_upper_estimate": 0.005322497685029183
     },
     {
       "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": 9709.000920867997,
-      "required_pages_per_second_upper_estimate": 0.004064955938573073
+      "required_token_positions_per_second": 9616.79526726443,
+      "required_pages_per_second_upper_estimate": 0.00402635135688226
     },
     {
       "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": 19418.001841735993,
-      "required_pages_per_second_upper_estimate": 0.004044257696488672
+      "required_token_positions_per_second": 19233.59053452886,
+      "required_pages_per_second_upper_estimate": 0.0040058496844506365
     },
     {
       "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": 29127.002762603992,
-      "required_pages_per_second_upper_estimate": 0.0017461193412600397
+      "required_token_positions_per_second": 28850.385801793287,
+      "required_pages_per_second_upper_estimate": 0.001729536576829081
     },
     {
       "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": 29127.002762603992,
-      "required_pages_per_second_upper_estimate": 0.002264496348647039
+      "required_token_positions_per_second": 28850.385801793287,
+      "required_pages_per_second_upper_estimate": 0.002242990596653432
     },
     {
       "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": 29127.002762603992,
-      "required_pages_per_second_upper_estimate": 0.3736099172997267
+      "required_token_positions_per_second": 28850.385801793287,
+      "required_pages_per_second_upper_estimate": 0.37006177193459916
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.06917773374229341,
-    "producer_p95_seconds": 0.23121268099930603,
-    "gpu_update_seconds": 0.9492222809652466,
-    "producer_batches_per_second": 14.455518356893307,
-    "gpu_batches_per_second": 1.0534940235316836,
-    "producer_headroom": 13.72150010726526,
+    "producer_batch_seconds": 0.12017947155473507,
+    "producer_p95_seconds": 0.2565946350005106,
+    "gpu_update_seconds": 0.9583234064856578,
+    "producer_batches_per_second": 8.320888643153632,
+    "gpu_batches_per_second": 1.0434890697986576,
+    "producer_headroom": 7.974102349494812,
     "sustainable": true,
-    "buffer_seconds": 60.750225981775785,
+    "buffer_seconds": 60.37437460859644,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 244524.96429285797,
-    "upload_bytes_per_second": 5066121.964446988,
+    "generation_bytes_per_second": 239961.1397314108,
+    "upload_bytes_per_second": 5071241.847889318,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.469365087407155,
+    "estimated_full_shards_per_hour": 51.490074576919014,
     "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": 1.3950513529998716,
-      "bytes": 7312918,
-      "pages": 32,
-      "shards": 3
-    },
     {
       "seconds": 1.2647196729994903,
       "bytes": 6433157,
@@ -562,6 +556,12 @@
       "bytes": 6180333,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.1313278150009864,
+      "bytes": 6410101,
+      "pages": 20,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -186,3 +186,4 @@
 {"step": 903600, "loss": 0.5156925916671753, "lr": 3e-05, "grad_norm": 0.32579942934665634, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132260982752, "tokens_per_second": 151680.22466240288, "data_wait_s": 7.382001058431342e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 903700, "loss": 0.5315797328948975, "lr": 3e-05, "grad_norm": 0.31796979437313505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132275494645, "tokens_per_second": 152303.42022954687, "data_wait_s": 1.2202999641885981e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 903800, "loss": 0.5106819272041321, "lr": 3e-05, "grad_norm": 0.3428163658388574, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132290056936, "tokens_per_second": 129264.76447701133, "data_wait_s": 9.261999366572127e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 903900, "loss": 0.5374611020088196, "lr": 3e-05, "grad_norm": 0.33639597405785376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 132304702741, "tokens_per_second": 151373.51801883074, "data_wait_s": 1.7398000636603683e-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:dd718053f653c9674f37632621029cc0c69c9d6d8bf8361186030a78b8ca926d
-size 400698
+oid sha256:91f51e08579c57df7862a16eff9d7d1e7e6f0dd91da42046c3cb26304bda3bdc
+size 402816