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": 76684.67312246532,
-      "required_pages_per_second_upper_estimate": 0.005305223619568172
+      "required_token_positions_per_second": 76903.0734224438,
+      "required_pages_per_second_upper_estimate": 0.005320333059079197
     },
     {
       "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": 9585.584140308165,
-      "required_pages_per_second_upper_estimate": 0.004013283909788117
+      "required_token_positions_per_second": 9612.884177805476,
+      "required_pages_per_second_upper_estimate": 0.004024713865398631
     },
     {
       "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": 19171.16828061633,
-      "required_pages_per_second_upper_estimate": 0.003992848775146193
+      "required_token_positions_per_second": 19225.76835561095,
+      "required_pages_per_second_upper_estimate": 0.0040042205308667665
     },
     {
       "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": 28756.752420924495,
-      "required_pages_per_second_upper_estimate": 0.0017239233986158914
+      "required_token_positions_per_second": 28838.652533416425,
+      "required_pages_per_second_upper_estimate": 0.0017288331853055517
     },
     {
       "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": 28756.752420924495,
-      "required_pages_per_second_upper_estimate": 0.0022357110131406
+      "required_token_positions_per_second": 28838.652533416425,
+      "required_pages_per_second_upper_estimate": 0.002242078386646324
     },
     {
       "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": 28756.752420924495,
-      "required_pages_per_second_upper_estimate": 0.36886074346050585
+      "required_token_positions_per_second": 28838.652533416425,
+      "required_pages_per_second_upper_estimate": 0.36991127016606284
     }
   ],
   "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": 0.07936180703106288,
-    "producer_p95_seconds": 0.30083170799844083,
-    "gpu_update_seconds": 0.961443753985317,
-    "producer_batches_per_second": 12.600519537169706,
-    "gpu_batches_per_second": 1.040102445779966,
-    "producer_headroom": 12.114690805981772,
+    "producer_batch_seconds": 0.19450978175771638,
+    "producer_p95_seconds": 0.31630176899852813,
+    "gpu_update_seconds": 0.9587133090897096,
+    "producer_batches_per_second": 5.141129618075514,
+    "gpu_batches_per_second": 1.0430646894320177,
+    "producer_headroom": 4.928869388604291,
     "sustainable": true,
-    "buffer_seconds": 60.570956501074974,
+    "buffer_seconds": 60.39893847265171,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 445479.4978759256,
-    "upload_bytes_per_second": 5166507.223510661,
+    "generation_bytes_per_second": 468207.31244441355,
+    "upload_bytes_per_second": 5166024.008308548,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 95.58953001221015,
+    "estimated_full_shards_per_hour": 100.46638994216256,
     "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": 1.2498522079986287,
-      "bytes": 10921992,
-      "pages": 31,
-      "shards": 3
-    },
     {
       "seconds": 1.2837640949983324,
       "bytes": 7864429,
@@ -562,6 +556,12 @@
       "bytes": 1074791,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.7156042310016346,
+      "bytes": 13288348,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -266,3 +266,4 @@
 {"step": 631000, "loss": 0.5597079396247864, "lr": 0.0001, "grad_norm": 0.1661672466958877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92361540978, "tokens_per_second": 151997.73454559484, "data_wait_s": 1.2456999684218317e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 631100, "loss": 0.5607529878616333, "lr": 0.0001, "grad_norm": 0.17976467938504226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92376179478, "tokens_per_second": 151324.45741752646, "data_wait_s": 1.4661000022897497e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 631200, "loss": 0.5444761514663696, "lr": 0.0001, "grad_norm": 0.174329482413636, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92390897406, "tokens_per_second": 152248.52500276762, "data_wait_s": 1.2279000657144934e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 631300, "loss": 0.5055897235870361, "lr": 0.0001, "grad_norm": 0.1672872135139346, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92405616373, "tokens_per_second": 152557.24394906053, "data_wait_s": 1.5546000213362277e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:251e22b1d615d83b616eae94007ce5ce944076e39a146368a8b46da2c552d738
-size 572487
+oid sha256:fb811f4f5167e0007974990bfa124d805d6de9f5a3b8e6fef18a1f1d47a99f6d
+size 574605