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": 76737.44308904043,
-      "required_pages_per_second_upper_estimate": 0.00530887436829903
+      "required_token_positions_per_second": 76637.3117956056,
+      "required_pages_per_second_upper_estimate": 0.005301947053082598
     },
     {
       "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": 9592.180386130054,
-      "required_pages_per_second_upper_estimate": 0.004016045620168465
+      "required_token_positions_per_second": 9579.6639744507,
+      "required_pages_per_second_upper_estimate": 0.004010805259970717
     },
     {
       "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": 19184.36077226011,
-      "required_pages_per_second_upper_estimate": 0.003995596423246192
+      "required_token_positions_per_second": 19159.3279489014,
+      "required_pages_per_second_upper_estimate": 0.0039903827463005186
     },
     {
       "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": 28776.541158390162,
-      "required_pages_per_second_upper_estimate": 0.0017251097032113057
+      "required_token_positions_per_second": 28738.9919233521,
+      "required_pages_per_second_upper_estimate": 0.001722858683905136
     },
     {
       "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": 28776.541158390162,
-      "required_pages_per_second_upper_estimate": 0.002237249500437098
+      "required_token_positions_per_second": 28738.9919233521,
+      "required_pages_per_second_upper_estimate": 0.0022343302125745193
     },
     {
       "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": 28776.541158390162,
-      "required_pages_per_second_upper_estimate": 0.36911457213722454
+      "required_token_positions_per_second": 28738.9919233521,
+      "required_pages_per_second_upper_estimate": 0.3686329308673837
     }
   ],
   "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": 0.061263879781222386,
-    "producer_p95_seconds": 0.2016101230001368,
-    "gpu_update_seconds": 0.9607825988474948,
-    "producer_batches_per_second": 16.322831717009603,
-    "gpu_batches_per_second": 1.0408181842589035,
-    "producer_headroom": 15.682692677618801,
+    "producer_batch_seconds": 0.08636832655469107,
+    "producer_p95_seconds": 0.21504635000019334,
+    "gpu_update_seconds": 0.9620379195532741,
+    "producer_batches_per_second": 11.578318579169986,
+    "gpu_batches_per_second": 1.0394600666721678,
+    "producer_headroom": 11.138781517829713,
     "sustainable": true,
-    "buffer_seconds": 60.529303727392175,
+    "buffer_seconds": 60.60838893185627,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369825.46943647787,
-    "upload_bytes_per_second": 5274766.173745846,
+    "generation_bytes_per_second": 365126.2115834569,
+    "upload_bytes_per_second": 5293858.481099821,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.35593664475205,
+    "estimated_full_shards_per_hour": 78.34758530261784,
     "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": 1.3394989830003396,
-      "bytes": 6035169,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.5462701479991665,
       "bytes": 6440094,
@@ -562,6 +556,12 @@
       "bytes": 6070099,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.0409215150011732,
+      "bytes": 6007655,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -126,3 +126,4 @@
 {"step": 657600, "loss": 0.5721409320831299, "lr": 0.0001, "grad_norm": 0.23453476625534417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96254934317, "tokens_per_second": 153055.72192716855, "data_wait_s": 9.101000614464283e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 657700, "loss": 0.576458215713501, "lr": 0.0001, "grad_norm": 0.20272162505566665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96269581173, "tokens_per_second": 152844.57654697844, "data_wait_s": 1.2495000191847794e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 657800, "loss": 0.5763843655586243, "lr": 0.0001, "grad_norm": 0.1878534780393081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96284144554, "tokens_per_second": 151669.88631891724, "data_wait_s": 1.0154000847251154e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 657900, "loss": 0.5477266311645508, "lr": 0.0001, "grad_norm": 0.20004618513616076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 96298802646, "tokens_per_second": 153721.23076655628, "data_wait_s": 1.2876000255346298e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:697e7dd4bd46a6929164958131a18fa41e4ac366c1c79df618f96c0ddf61d752
-size 272835
+oid sha256:be7ccde18593975fa02e82cf98cd33c1731b1dd6d1695459a600a9997c629854
+size 274953