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": 77288.01888292708,
-      "required_pages_per_second_upper_estimate": 0.005346964479231957
+      "required_token_positions_per_second": 77121.74074816429,
+      "required_pages_per_second_upper_estimate": 0.005335460972050639
     },
     {
       "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": 9661.002360365885,
-      "required_pages_per_second_upper_estimate": 0.004044859943614761
+      "required_token_positions_per_second": 9640.217593520536,
+      "required_pages_per_second_upper_estimate": 0.0040361577957719
     },
     {
       "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": 19322.00472073177,
-      "required_pages_per_second_upper_estimate": 0.004024264027797817
+      "required_token_positions_per_second": 19280.435187041072,
+      "required_pages_per_second_upper_estimate": 0.0040156061901923685
     },
     {
       "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": 28983.00708109765,
-      "required_pages_per_second_upper_estimate": 0.0017374870200224093
+      "required_token_positions_per_second": 28920.65278056161,
+      "required_pages_per_second_upper_estimate": 0.0017337489749147778
     },
     {
       "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": 28983.00708109765,
-      "required_pages_per_second_upper_estimate": 0.0022533013177799877
+      "required_token_positions_per_second": 28920.65278056161,
+      "required_pages_per_second_upper_estimate": 0.002248453545180893
     },
     {
       "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": 28983.00708109765,
-      "required_pages_per_second_upper_estimate": 0.3717628953078802
+      "required_token_positions_per_second": 28920.65278056161,
+      "required_pages_per_second_upper_estimate": 0.37096308129143557
     }
   ],
   "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": 0.08058367901571728,
-    "producer_p95_seconds": 0.23394188699967344,
-    "gpu_update_seconds": 0.9539382826163567,
-    "producer_batches_per_second": 12.409460727214466,
-    "gpu_batches_per_second": 1.0482858463938676,
-    "producer_headroom": 11.837859654314093,
+    "producer_batch_seconds": 0.30567738092186403,
+    "producer_p95_seconds": 0.22256744299920683,
+    "gpu_update_seconds": 0.9559950188462898,
+    "producer_batches_per_second": 3.2714229524742486,
+    "gpu_batches_per_second": 1.0460305548524889,
+    "producer_headroom": 3.1274640471048043,
     "sustainable": true,
-    "buffer_seconds": 60.098111804830474,
+    "buffer_seconds": 60.22768618731626,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 440679.5180489529,
-    "upload_bytes_per_second": 5295343.635228061,
+    "generation_bytes_per_second": 392640.2820072295,
+    "upload_bytes_per_second": 5319385.265754297,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.55956607915344,
+    "estimated_full_shards_per_hour": 84.25146432078041,
     "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": 1.123353902999952,
-      "bytes": 7475982,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.3653379679999489,
       "bytes": 11617938,
@@ -562,6 +556,12 @@
       "bytes": 1226986,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.5480925499996374,
+      "bytes": 11721285,
+      "pages": 32,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -138,3 +138,4 @@
 {"step": 898800, "loss": 0.5405122637748718, "lr": 3e-05, "grad_norm": 0.3397679802501427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131558747977, "tokens_per_second": 150542.24266902485, "data_wait_s": 1.3663999197888188e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 898900, "loss": 0.5080359578132629, "lr": 3e-05, "grad_norm": 0.3221479352108406, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131573286793, "tokens_per_second": 151457.5214620092, "data_wait_s": 1.1776000974350609e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 899000, "loss": 0.5392065048217773, "lr": 3e-05, "grad_norm": 0.31953209477209304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131587952390, "tokens_per_second": 152952.31749392705, "data_wait_s": 1.785200038284529e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 899100, "loss": 0.519760251045227, "lr": 3e-05, "grad_norm": 0.3299807719404503, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131602603846, "tokens_per_second": 153163.8475348971, "data_wait_s": 1.1886999345733784e-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 sha256:35f945623f994c4d7cc24f0d694b2d8c9d8f86703aac4825da19ef2237ee512e
-size 298251
+oid sha256:ca89c1d96a94c366b1a099db2c9e29a8ed474c39eb894e62c4c912b796457c07
+size 300369