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": 77635.16602999288,
-      "required_pages_per_second_upper_estimate": 0.005370980924358325
+      "required_token_positions_per_second": 77760.82570911407,
+      "required_pages_per_second_upper_estimate": 0.005379674352530556
     },
     {
       "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": 9704.39575374911,
-      "required_pages_per_second_upper_estimate": 0.00406302785126722
+      "required_token_positions_per_second": 9720.103213639259,
+      "required_pages_per_second_upper_estimate": 0.004069604236714164
     },
     {
       "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": 19408.79150749822,
-      "required_pages_per_second_upper_estimate": 0.004042339426759791
+      "required_token_positions_per_second": 19440.206427278517,
+      "required_pages_per_second_upper_estimate": 0.0040488823260828805
     },
     {
       "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": 29113.187261247327,
-      "required_pages_per_second_upper_estimate": 0.0017452911230487567
+      "required_token_positions_per_second": 29160.309640917778,
+      "required_pages_per_second_upper_estimate": 0.0017481160377582927
     },
     {
       "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": 29113.187261247327,
-      "required_pages_per_second_upper_estimate": 0.002263422254184527
+      "required_token_positions_per_second": 29160.309640917778,
+      "required_pages_per_second_upper_estimate": 0.0022670858119344614
     },
     {
       "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": 29113.187261247327,
-      "required_pages_per_second_upper_estimate": 0.373432706882253
+      "required_token_positions_per_second": 29160.309640917778,
+      "required_pages_per_second_upper_estimate": 0.37403714217259626
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10898179265666386,
-    "producer_p95_seconds": 0.32128083999850787,
-    "gpu_update_seconds": 0.9496727291278876,
-    "producer_batches_per_second": 9.175844658294428,
-    "gpu_batches_per_second": 1.052994330918957,
-    "producer_headroom": 8.71404943869602,
+    "producer_batch_seconds": 0.089136461898363,
+    "producer_p95_seconds": 0.3178858260071138,
+    "gpu_update_seconds": 0.9481380801664842,
+    "producer_batches_per_second": 11.218753568435782,
+    "gpu_batches_per_second": 1.054698699396621,
+    "producer_headroom": 10.636927470237596,
     "sustainable": true,
-    "buffer_seconds": 60.779054664184805,
+    "buffer_seconds": 60.680837130654986,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 260651.8064478773,
-    "upload_bytes_per_second": 5110944.552348715,
+    "generation_bytes_per_second": 274549.6959126535,
+    "upload_bytes_per_second": 5045332.30999754,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 55.92981000020255,
+    "estimated_full_shards_per_hour": 58.91197355303482,
     "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": 1.2456738490000134,
-      "bytes": 7708205,
-      "pages": 43,
-      "shards": 3
-    },
     {
       "seconds": 1.1626391710015014,
       "bytes": 1209462,
@@ -562,6 +556,12 @@
       "bytes": 7055090,
       "pages": 43,
       "shards": 3
+    },
+    {
+      "seconds": 2.0782521860019187,
+      "bytes": 6362598,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -406,3 +406,4 @@
 {"step": 925600, "loss": 0.5387578010559082, "lr": 3e-05, "grad_norm": 0.33788941470807277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135481383545, "tokens_per_second": 151936.79056625575, "data_wait_s": 7.653005013708025e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 925700, "loss": 0.5607717037200928, "lr": 3e-05, "grad_norm": 0.30635413400909806, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135496029460, "tokens_per_second": 153080.14015474648, "data_wait_s": 1.4951998309697956e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 925800, "loss": 0.5568497776985168, "lr": 3e-05, "grad_norm": 0.326474093277394, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135510638316, "tokens_per_second": 153377.92235050412, "data_wait_s": 1.788899680832401e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 925900, "loss": 0.5447126626968384, "lr": 3e-05, "grad_norm": 0.35461924083772356, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 135525342509, "tokens_per_second": 153940.6244063128, "data_wait_s": 8.148002962116152e-06, "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 sha256:75a7357e7d239ea61c113bb59d6bb9f6a6fd4bcae8ee6b1c87dec1cd09bc7f18
-size 870573
+oid sha256:7453d908e4e10968512724065e43b0425849cb04d013bfd046aa61127e4ac063
+size 872691