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": 76988.64664380987,
-      "required_pages_per_second_upper_estimate": 0.005326253213090532
+      "required_token_positions_per_second": 76709.9058357439,
+      "required_pages_per_second_upper_estimate": 0.005306969277220746
     },
     {
       "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": 9623.580830476234,
-      "required_pages_per_second_upper_estimate": 0.004029192330500369
+      "required_token_positions_per_second": 9588.738229467988,
+      "required_pages_per_second_upper_estimate": 0.0040146044610544634
     },
     {
       "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": 19247.161660952468,
-      "required_pages_per_second_upper_estimate": 0.004008676192190003
+      "required_token_positions_per_second": 19177.476458935977,
+      "required_pages_per_second_upper_estimate": 0.003994162602332326
     },
     {
       "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": 28870.7424914287,
-      "required_pages_per_second_upper_estimate": 0.0017307569292898588
+      "required_token_positions_per_second": 28766.214688403965,
+      "required_pages_per_second_upper_estimate": 0.0017244906471032397
     },
     {
       "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": 28870.7424914287,
-      "required_pages_per_second_upper_estimate": 0.00224457323973297
+      "required_token_positions_per_second": 28766.214688403965,
+      "required_pages_per_second_upper_estimate": 0.0022364466628170127
     },
     {
       "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": 28870.7424914287,
-      "required_pages_per_second_upper_estimate": 0.3703228856919319
+      "required_token_positions_per_second": 28766.214688403965,
+      "required_pages_per_second_upper_estimate": 0.3689821152679412
     }
   ],
   "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": 0.0872124334921125,
-    "producer_p95_seconds": 0.2508970019989647,
-    "gpu_update_seconds": 0.9576476950050137,
-    "producer_batches_per_second": 11.466254981756014,
-    "gpu_batches_per_second": 1.0442253505291053,
-    "producer_headroom": 10.980632653618402,
+    "producer_batch_seconds": 0.1293031230392785,
+    "producer_p95_seconds": 0.2686524649980129,
+    "gpu_update_seconds": 0.9611274997243648,
+    "producer_batches_per_second": 7.733765252493007,
+    "gpu_batches_per_second": 1.0404446863572034,
+    "producer_headroom": 7.4331344605837755,
     "sustainable": true,
-    "buffer_seconds": 60.33180478531586,
+    "buffer_seconds": 60.55103248263499,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 301029.86325652606,
-    "upload_bytes_per_second": 4976680.001449399,
+    "generation_bytes_per_second": 341238.70555039064,
+    "upload_bytes_per_second": 4979815.74556471,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.59400103828274,
+    "estimated_full_shards_per_hour": 73.22188258060254,
     "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": 1.195207675000347,
-      "bytes": 1134238,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.0769621990002634,
       "bytes": 6468284,
@@ -562,6 +556,12 @@
       "bytes": 5785840,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.1340223679981136,
+      "bytes": 1097457,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -196,3 +196,4 @@
 {"step": 779600, "loss": 0.5156227946281433, "lr": 3e-05, "grad_norm": 0.26047371955239, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114111758568, "tokens_per_second": 151882.10459962077, "data_wait_s": 1.2558000889839604e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 779700, "loss": 0.49602168798446655, "lr": 3e-05, "grad_norm": 0.2439595266866311, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114126194413, "tokens_per_second": 150062.33628573152, "data_wait_s": 1.2019998393952847e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 779800, "loss": 0.5133309364318848, "lr": 3e-05, "grad_norm": 0.24772370564504653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114140554045, "tokens_per_second": 148428.1273339138, "data_wait_s": 1.1033000191673636e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 779900, "loss": 0.521408200263977, "lr": 3e-05, "grad_norm": 0.239432013755575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 114155151564, "tokens_per_second": 151979.67729621593, "data_wait_s": 1.020299896481447e-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:bd76226bc4f7e8e71db556b59e7513e6f3ce8ae0a1f78983a5860728e8c3f754
-size 421878
+oid sha256:cf2d90186c2c166c34c317932b6b4415944a0da4614a6068ecbd7349eb9ca303
+size 423996