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": 76652.30968938455,
-      "required_pages_per_second_upper_estimate": 0.005302984642174134
+      "required_token_positions_per_second": 76526.41261864408,
+      "required_pages_per_second_upper_estimate": 0.005294274790698858
     },
     {
       "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": 9581.538711173069,
-      "required_pages_per_second_upper_estimate": 0.004011590173087419
+      "required_token_positions_per_second": 9565.80157733051,
+      "required_pages_per_second_upper_estimate": 0.004005001363776264
     },
     {
       "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": 19163.077422346138,
-      "required_pages_per_second_upper_estimate": 0.003991163662738793
+      "required_token_positions_per_second": 19131.60315466102,
+      "required_pages_per_second_upper_estimate": 0.00398460840281224
     },
     {
       "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": 28744.61613351921,
-      "required_pages_per_second_upper_estimate": 0.0017231958467169824
+      "required_token_positions_per_second": 28697.404731991533,
+      "required_pages_per_second_upper_estimate": 0.0017203655952830344
     },
     {
       "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": 28744.61613351921,
-      "required_pages_per_second_upper_estimate": 0.0022347674701767256
+      "required_token_positions_per_second": 28697.404731991533,
+      "required_pages_per_second_upper_estimate": 0.002231096991368958
     },
     {
       "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": 28744.61613351921,
-      "required_pages_per_second_upper_estimate": 0.368705072196601
+      "required_token_positions_per_second": 28697.404731991533,
+      "required_pages_per_second_upper_estimate": 0.3680994950294575
     }
   ],
   "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": 0.07091389871087017,
-    "producer_p95_seconds": 0.24715099099921645,
-    "gpu_update_seconds": 0.9618496859229079,
-    "producer_batches_per_second": 14.101607980647003,
-    "gpu_batches_per_second": 1.03966348862555,
-    "producer_headroom": 13.56362720719329,
+    "producer_batch_seconds": 0.0908051666953611,
+    "producer_p95_seconds": 0.24614899399966816,
+    "gpu_update_seconds": 0.963432068446884,
+    "producer_batches_per_second": 11.012589221435638,
+    "gpu_batches_per_second": 1.0379559003179808,
+    "producer_headroom": 10.609881612563598,
     "sustainable": true,
-    "buffer_seconds": 60.5965302131432,
+    "buffer_seconds": 60.69622031215369,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323357.00722155283,
-    "upload_bytes_per_second": 5054679.604209175,
+    "generation_bytes_per_second": 324092.3744192408,
+    "upload_bytes_per_second": 5045159.515345617,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.38488638386669,
+    "estimated_full_shards_per_hour": 69.54267906601828,
     "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": 1.5044031500001438,
-      "bytes": 8199304,
-      "pages": 34,
-      "shards": 3
-    },
     {
       "seconds": 1.4808365860008053,
       "bytes": 6083362,
@@ -562,6 +556,12 @@
       "bytes": 6256652,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3504138789976423,
+      "bytes": 6609036,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -186,3 +186,4 @@
 {"step": 778600, "loss": 0.537959098815918, "lr": 3e-05, "grad_norm": 0.25468201989668166, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113965021281, "tokens_per_second": 150455.7726355637, "data_wait_s": 1.4880999515298754e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 778700, "loss": 0.5149663090705872, "lr": 3e-05, "grad_norm": 0.2534125750871316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113979663185, "tokens_per_second": 151606.67270118775, "data_wait_s": 1.4108001778367907e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 778800, "loss": 0.5007060170173645, "lr": 3e-05, "grad_norm": 0.25200083669656076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 113994176375, "tokens_per_second": 150618.73996273061, "data_wait_s": 1.556899951538071e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 778900, "loss": 0.5157540440559387, "lr": 3e-05, "grad_norm": 0.24314472441481658, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114008826461, "tokens_per_second": 153012.47914859478, "data_wait_s": 1.2923999747727066e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:47b3f9dc21b733fb7a1e660f97e5f844fcc55bbe38baae6888aa242ca81f1b32
-size 400698
+oid sha256:47041e268d33caf8831616673c714614fa887b3c6ea1e42acfb4096ac1c71ceb
+size 402816