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": 77492.71061061385,
-      "required_pages_per_second_upper_estimate": 0.005361125527903572
+      "required_token_positions_per_second": 77322.00500718379,
+      "required_pages_per_second_upper_estimate": 0.00534931571816671
     },
     {
       "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": 9686.588826326732,
-      "required_pages_per_second_upper_estimate": 0.0040555724626064
+      "required_token_positions_per_second": 9665.250625897974,
+      "required_pages_per_second_upper_estimate": 0.0040466386036791735
     },
     {
       "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": 19373.177652653463,
-      "required_pages_per_second_upper_estimate": 0.004034921999996139
+      "required_token_positions_per_second": 19330.501251795948,
+      "required_pages_per_second_upper_estimate": 0.004026033631149893
     },
     {
       "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": 29059.766478980193,
-      "required_pages_per_second_upper_estimate": 0.0017420886287206536
+      "required_token_positions_per_second": 28995.75187769392,
+      "required_pages_per_second_upper_estimate": 0.0017382510511181782
     },
     {
       "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": 29059.766478980193,
-      "required_pages_per_second_upper_estimate": 0.0022592690233364456
+      "required_token_positions_per_second": 28995.75187769392,
+      "required_pages_per_second_upper_estimate": 0.002254292169656912
     },
     {
       "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": 29059.766478980193,
-      "required_pages_per_second_upper_estimate": 0.3727474824460973
+      "required_token_positions_per_second": 28995.75187769392,
+      "required_pages_per_second_upper_estimate": 0.37192637187432076
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.11088183010155461,
-    "producer_p95_seconds": 0.18175020000035147,
-    "gpu_update_seconds": 0.9514185194846158,
-    "producer_batches_per_second": 9.018610164389589,
-    "gpu_batches_per_second": 1.0510621556344109,
-    "producer_headroom": 8.58047273041245,
+    "producer_batch_seconds": 0.4460236786093148,
+    "producer_p95_seconds": 0.19942733399966528,
+    "gpu_update_seconds": 0.9535189884580737,
+    "producer_batches_per_second": 2.2420334344534414,
+    "gpu_batches_per_second": 1.0487468127059434,
+    "producer_headroom": 2.1378214525092263,
     "sustainable": true,
-    "buffer_seconds": 60.89078524701541,
+    "buffer_seconds": 60.071696272858645,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 314407.2346590987,
-    "upload_bytes_per_second": 5197019.741949496,
+    "generation_bytes_per_second": 313464.15864199266,
+    "upload_bytes_per_second": 5289093.047335207,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.46447353200647,
+    "estimated_full_shards_per_hour": 67.2621113724216,
     "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": [
@@ -406,6 +406,12 @@
       "bytes": 7475982,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.3653379679999489,
+      "bytes": 11617938,
+      "pages": 31,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -58,3 +58,4 @@
 {"step": 890800, "loss": 0.5203015804290771, "lr": 3e-05, "grad_norm": 0.3147710947317427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130388333487, "tokens_per_second": 151918.67317398838, "data_wait_s": 1.0480999662831891e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 890900, "loss": 0.5388245582580566, "lr": 3e-05, "grad_norm": 0.3211956049093675, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130402871994, "tokens_per_second": 151829.41128575586, "data_wait_s": 1.174099998024758e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 891000, "loss": 0.5289643406867981, "lr": 3e-05, "grad_norm": 0.3304750209118958, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130417428723, "tokens_per_second": 137786.4988271708, "data_wait_s": 1.584900019224733e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 891100, "loss": 0.5342656970024109, "lr": 3e-05, "grad_norm": 0.3095160137590757, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 130432070713, "tokens_per_second": 152782.13955704786, "data_wait_s": 1.8817000636772718e-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 sha256:6aa67b9091c5f78e5e18b028d49162776103c415ea02345d22d61662679c2551
-size 128028
+oid sha256:89405cf7b635d437fa7ad83118b1c1e9183989158f7eeaf75d8a8b6acde3b1ea
+size 130146