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": 76222.21905006251,
-      "required_pages_per_second_upper_estimate": 0.005273229973798062
+      "required_token_positions_per_second": 76940.27896482416,
+      "required_pages_per_second_upper_estimate": 0.005322907024829822
     },
     {
       "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": 9527.777381257814,
-      "required_pages_per_second_upper_estimate": 0.003989081426916133
+      "required_token_positions_per_second": 9617.53487060302,
+      "required_pages_per_second_upper_estimate": 0.004026661013355454
     },
     {
       "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": 19055.55476251563,
-      "required_pages_per_second_upper_estimate": 0.003968769528259321
+      "required_token_positions_per_second": 19235.06974120604,
+      "required_pages_per_second_upper_estimate": 0.00400615776419219
     },
     {
       "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": 28583.33214377344,
-      "required_pages_per_second_upper_estimate": 0.0017135271177981166
+      "required_token_positions_per_second": 28852.604611809063,
+      "required_pages_per_second_upper_estimate": 0.0017296695911016003
     },
     {
       "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": 28583.33214377344,
-      "required_pages_per_second_upper_estimate": 0.00222222834938729
+      "required_token_positions_per_second": 28852.604611809063,
+      "required_pages_per_second_upper_estimate": 0.0022431630993726445
     },
     {
       "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": 28583.33214377344,
-      "required_pages_per_second_upper_estimate": 0.3666362943493983
+      "required_token_positions_per_second": 28852.604611809063,
+      "required_pages_per_second_upper_estimate": 0.370090232447109
     }
   ],
   "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": 0.14710983403881528,
-    "producer_p95_seconds": 0.3817793120033457,
-    "gpu_update_seconds": 0.9672770081854437,
-    "producer_batches_per_second": 6.797642091936203,
-    "gpu_batches_per_second": 1.033830010987176,
-    "producer_headroom": 6.575202905403492,
+    "producer_batch_seconds": 0.11141242660943362,
+    "producer_p95_seconds": 0.34462235800310737,
+    "gpu_update_seconds": 0.9582497099303114,
+    "producer_batches_per_second": 8.975659452294229,
+    "gpu_batches_per_second": 1.0435693218970292,
+    "producer_headroom": 8.600923066594204,
     "sustainable": true,
-    "buffer_seconds": 60.93845151568296,
+    "buffer_seconds": 60.36973172560962,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 76,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 276386.3809488254,
-    "upload_bytes_per_second": 4349343.853216611,
+    "generation_bytes_per_second": 265394.22156089876,
+    "upload_bytes_per_second": 4369922.811838172,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.3060834059579,
+    "estimated_full_shards_per_hour": 56.94742188568327,
     "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": 1.5883049809999648,
-      "bytes": 5665128,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 2.0745964650050155,
       "bytes": 11457524,
@@ -562,6 +556,12 @@
       "bytes": 12133874,
       "pages": 26,
       "shards": 4
+    },
+    {
+      "seconds": 1.269238109998696,
+      "bytes": 6277214,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -414,3 +414,4 @@
 {"step": 841400, "loss": 0.5326145887374878, "lr": 3e-05, "grad_norm": 0.30268375799144104, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123156957077, "tokens_per_second": 153413.4737340514, "data_wait_s": 7.086004188749939e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 841500, "loss": 0.5076887011528015, "lr": 3e-05, "grad_norm": 0.29050081810129474, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123171816266, "tokens_per_second": 105324.0332797843, "data_wait_s": 1.6270001651719213e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 841600, "loss": 0.5214629173278809, "lr": 3e-05, "grad_norm": 0.2913366437421171, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123186460832, "tokens_per_second": 151638.82484134458, "data_wait_s": 9.404000593349338e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 841700, "loss": 0.5699270963668823, "lr": 3e-05, "grad_norm": 0.29822395803538615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 123201045057, "tokens_per_second": 149281.0584494862, "data_wait_s": 1.1142998118884861e-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:d111be122445c5ea63c57fac78764b0014e56e88f32af3e590999926473469cc
-size 887517
+oid sha256:09f2820cca8cafb104b060aad0ed1482ef725ccf5b3105cf94752fa77d7ce278
+size 889635