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": 77759.74997250341,
-      "required_pages_per_second_upper_estimate": 0.0053795999305757776
+      "required_token_positions_per_second": 77107.19785918115,
+      "required_pages_per_second_upper_estimate": 0.0053344548612466345
     },
     {
       "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": 9719.968746562927,
-      "required_pages_per_second_upper_estimate": 0.0040695479381572955
+      "required_token_positions_per_second": 9638.399732397644,
+      "required_pages_per_second_upper_estimate": 0.004035396695281002
     },
     {
       "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": 19439.937493125854,
-      "required_pages_per_second_upper_estimate": 0.0040488263141911516
+      "required_token_positions_per_second": 19276.799464795287,
+      "required_pages_per_second_upper_estimate": 0.004014848965129015
     },
     {
       "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": 29159.90623968878,
-      "required_pages_per_second_upper_estimate": 0.0017480918544705722
+      "required_token_positions_per_second": 28915.19919719293,
+      "required_pages_per_second_upper_estimate": 0.0017334220409189651
     },
     {
       "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": 29159.90623968878,
-      "required_pages_per_second_upper_estimate": 0.002267054449263281
+      "required_token_positions_per_second": 28915.19919719293,
+      "required_pages_per_second_upper_estimate": 0.0022480295530617565
     },
     {
       "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": 29159.90623968878,
-      "required_pages_per_second_upper_estimate": 0.37403196777476916
+      "required_token_positions_per_second": 28915.19919719293,
+      "required_pages_per_second_upper_estimate": 0.3708931285795838
     }
   ],
   "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": 0.09001878137490849,
-    "producer_p95_seconds": 0.3150593879981898,
-    "gpu_update_seconds": 0.9481511968090294,
-    "producer_batches_per_second": 11.108792906618223,
-    "gpu_batches_per_second": 1.0546841087850398,
-    "producer_headroom": 10.532815289513724,
+    "producer_batch_seconds": 0.23448007978862506,
+    "producer_p95_seconds": 0.3324978890013881,
+    "gpu_update_seconds": 0.9561753253522131,
+    "producer_batches_per_second": 4.264754604747074,
+    "gpu_batches_per_second": 1.0458333042966192,
+    "producer_headroom": 4.0778531217413825,
     "sustainable": true,
-    "buffer_seconds": 60.68167659577788,
+    "buffer_seconds": 60.23904549718942,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 236678.19462116563,
-    "upload_bytes_per_second": 5014414.103612541,
+    "generation_bytes_per_second": 268110.18415177404,
+    "upload_bytes_per_second": 5054344.713210194,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 50.78563097931124,
+    "estimated_full_shards_per_hour": 57.53020423331181,
     "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": 1.1096691220009234,
-      "bytes": 969203,
-      "pages": 31,
-      "shards": 1
-    },
     {
       "seconds": 1.6109700509987306,
       "bytes": 10483727,
@@ -562,6 +556,12 @@
       "bytes": 6356822,
       "pages": 43,
       "shards": 2
+    },
+    {
+      "seconds": 1.7031787460000487,
+      "bytes": 7399641,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -413,3 +413,4 @@
 {"step": 926300, "loss": 0.5263802409172058, "lr": 3e-05, "grad_norm": 0.3582380577863063, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135584278242, "tokens_per_second": 153598.82214552615, "data_wait_s": 1.4899997040629387e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 926400, "loss": 0.6199176907539368, "lr": 3e-05, "grad_norm": 0.36822221233973507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135598825732, "tokens_per_second": 152560.8380627182, "data_wait_s": 1.2646996765397489e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 926500, "loss": 0.533126711845398, "lr": 3e-05, "grad_norm": 0.3618404290064263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135613285555, "tokens_per_second": 151126.4694306173, "data_wait_s": 1.1654999980237335e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 926600, "loss": 0.504806399345398, "lr": 3e-05, "grad_norm": 0.3150711242582432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135627853641, "tokens_per_second": 152076.48591215358, "data_wait_s": 1.6559002688154578e-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 sha256:7040dce8c4e83c77b806cf4fe5bfefa0548fba8d6708a85c8ddb5902a80cdbd3
-size 885399
+oid sha256:f936d5c2542391ecf227b81f91e08adc6496cd19dd60edf06a55e188e8dc2477
+size 887517