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": 76367.44879080764,
-      "required_pages_per_second_upper_estimate": 0.005283277304242235
+      "required_token_positions_per_second": 77101.9359975511,
+      "required_pages_per_second_upper_estimate": 0.005334090833449867
     },
     {
       "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": 9545.931098850955,
-      "required_pages_per_second_upper_estimate": 0.0039966820093796495
+      "required_token_positions_per_second": 9637.741999693888,
+      "required_pages_per_second_upper_estimate": 0.004035121316332961
     },
     {
       "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": 19091.86219770191,
-      "required_pages_per_second_upper_estimate": 0.00397633140951717
+      "required_token_positions_per_second": 19275.483999387776,
+      "required_pages_per_second_upper_estimate": 0.004014574988375786
     },
     {
       "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": 28637.793296552863,
-      "required_pages_per_second_upper_estimate": 0.0017167919807498776
+      "required_token_positions_per_second": 28913.225999081664,
+      "required_pages_per_second_upper_estimate": 0.0017333037507050938
     },
     {
       "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": 28637.793296552863,
-      "required_pages_per_second_upper_estimate": 0.0022264624644665956
+      "required_token_positions_per_second": 28913.225999081664,
+      "required_pages_per_second_upper_estimate": 0.0022478761455877864
     },
     {
       "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": 28637.793296552863,
-      "required_pages_per_second_upper_estimate": 0.3673348635414228
+      "required_token_positions_per_second": 28913.225999081664,
+      "required_pages_per_second_upper_estimate": 0.3708678185128675
     }
   ],
   "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": 0.32668154600793287,
-    "producer_p95_seconds": 0.3290660509992449,
-    "gpu_update_seconds": 0.9654375151638515,
-    "producer_batches_per_second": 3.0610850604206363,
-    "gpu_batches_per_second": 1.035799815413515,
-    "producer_headroom": 2.955286354437687,
+    "producer_batch_seconds": 0.10911405300777233,
+    "producer_p95_seconds": 0.30382779800129356,
+    "gpu_update_seconds": 0.9562405800334474,
+    "producer_batches_per_second": 9.164722347255937,
+    "gpu_batches_per_second": 1.0457619357306736,
+    "producer_headroom": 8.763679413185516,
     "sustainable": true,
-    "buffer_seconds": 60.822563455322644,
+    "buffer_seconds": 60.24315654210719,
     "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": 498361.7194116791,
-    "upload_bytes_per_second": 5070954.374711999,
+    "generation_bytes_per_second": 497944.1343682172,
+    "upload_bytes_per_second": 5136994.879520459,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.9368237186697,
+    "estimated_full_shards_per_hour": 106.84721968922507,
     "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": 1.0001321140007349,
-      "bytes": 5054764,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.5159087600004568,
       "bytes": 7203777,
@@ -562,6 +556,12 @@
       "bytes": 865962,
       "pages": 27,
       "shards": 1
+    },
+    {
+      "seconds": 1.0794645450005191,
+      "bytes": 10910956,
+      "pages": 37,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -273,3 +273,4 @@
 {"step": 631700, "loss": 0.5200865268707275, "lr": 0.0001, "grad_norm": 0.16134545844521925, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92464076764, "tokens_per_second": 150018.86339453675, "data_wait_s": 1.6152000171132386e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 631800, "loss": 0.49992790818214417, "lr": 0.0001, "grad_norm": 0.1781931082507836, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92478667705, "tokens_per_second": 150913.55551660372, "data_wait_s": 1.8184000509791076e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 631900, "loss": 0.5687424540519714, "lr": 0.0001, "grad_norm": 0.19371274338738628, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92493335730, "tokens_per_second": 152375.890852176, "data_wait_s": 1.7825997929321602e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 632000, "loss": 0.5255146622657776, "lr": 0.0001, "grad_norm": 0.16012158731824785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92507969610, "tokens_per_second": 151008.07078913908, "data_wait_s": 1.7130001651821658e-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:fbff69a963979d10ffd0898f619e6f65c27412b7343d92aa203790680f849988
-size 587313
+oid sha256:50b26ee727baf3ef71366880353ef0b019cd49906f3f1fbe6c1ba79565b81862
+size 589431