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": 76627.088230956,
-      "required_pages_per_second_upper_estimate": 0.0053012397631582
+      "required_token_positions_per_second": 76647.03978253974,
+      "required_pages_per_second_upper_estimate": 0.005302620057790741
     },
     {
       "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": 9578.3860288695,
-      "required_pages_per_second_upper_estimate": 0.004010270210842408
+      "required_token_positions_per_second": 9580.879972817467,
+      "required_pages_per_second_upper_estimate": 0.004011314373093949
     },
     {
       "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": 19156.772057739,
-      "required_pages_per_second_upper_estimate": 0.003989850421574773
+      "required_token_positions_per_second": 19161.759945634934,
+      "required_pages_per_second_upper_estimate": 0.003990889267084048
     },
     {
       "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": 28735.1580866085,
-      "required_pages_per_second_upper_estimate": 0.0017226288512462867
+      "required_token_positions_per_second": 28742.6399184524,
+      "required_pages_per_second_upper_estimate": 0.001723077375641233
     },
     {
       "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": 28735.1580866085,
-      "required_pages_per_second_upper_estimate": 0.00223403214863097
+      "required_token_positions_per_second": 28742.6399184524,
+      "required_pages_per_second_upper_estimate": 0.002234613828147733
     },
     {
       "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": 28735.1580866085,
-      "required_pages_per_second_upper_estimate": 0.36858375452608994
+      "required_token_positions_per_second": 28742.6399184524,
+      "required_pages_per_second_upper_estimate": 0.36867972343161837
     }
   ],
   "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": 0.21019920923444602,
-    "producer_p95_seconds": 0.2930866030001198,
-    "gpu_update_seconds": 0.9621662743830476,
-    "producer_batches_per_second": 4.757391826743974,
-    "gpu_batches_per_second": 1.0393214007019855,
-    "producer_headroom": 4.57740196971861,
+    "producer_batch_seconds": 0.07062829977343199,
+    "producer_p95_seconds": 0.2690986419984256,
+    "gpu_update_seconds": 0.9619158183953153,
+    "producer_batches_per_second": 14.158630509411847,
+    "gpu_batches_per_second": 1.0395920109393952,
+    "producer_headroom": 13.619410653817775,
     "sustainable": true,
-    "buffer_seconds": 60.616475286132,
+    "buffer_seconds": 60.600696558904865,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 661661.7980465048,
-    "upload_bytes_per_second": 3955082.50993938,
+    "generation_bytes_per_second": 660933.4768952767,
+    "upload_bytes_per_second": 4010835.9049177743,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.9886095812147,
+    "estimated_full_shards_per_hour": 70.91046919891227,
     "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": 1.8819817990006413,
-      "bytes": 6541422,
-      "pages": 23,
-      "shards": 2
-    },
     {
       "seconds": 1.735375592001219,
       "bytes": 12429197,
@@ -562,6 +556,12 @@
       "bytes": 2617112,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.5108418660020106,
+      "bytes": 11046966,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791300814123050567-72be911d/train.jsonl

@@ -236,3 +236,4 @@
 {"step": 823600, "loss": 0.5208242535591125, "lr": 3e-05, "grad_norm": 0.2910052313811568, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120551893798, "tokens_per_second": 151255.03495848106, "data_wait_s": 1.5976998838596046e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 823700, "loss": 0.5073413848876953, "lr": 3e-05, "grad_norm": 0.28283460045958064, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120566547181, "tokens_per_second": 151235.4149206287, "data_wait_s": 1.0415998985990882e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 823800, "loss": 0.5772274732589722, "lr": 3e-05, "grad_norm": 0.3110296014175407, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120581348880, "tokens_per_second": 152896.95934864332, "data_wait_s": 1.4068999007577077e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 823900, "loss": 0.5389016270637512, "lr": 3e-05, "grad_norm": 0.2831273121261325, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 120596101097, "tokens_per_second": 151948.56498266652, "data_wait_s": 1.629399775993079e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791300814.0c9d201b-2dbc-40fb-b739-4ee2350a0f54-7b2tx.840.0.1791300814123050567-72be911d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:59ad1f5daec23a836e6d37fb11b2fe04060a7ef9fd134a2d7dfa75288b4710c2
-size 507381
+oid sha256:673f1ebc5e0575e861868ee92c5562589bde15cf152d591fbc665e1389738a29
+size 509499