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": 76445.90440437887,
-      "required_pages_per_second_upper_estimate": 0.005288705045631197
+      "required_token_positions_per_second": 75003.82903529095,
+      "required_pages_per_second_upper_estimate": 0.005188938925521839
     },
     {
       "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": 9555.73805054736,
-      "required_pages_per_second_upper_estimate": 0.0040007879752624945
+      "required_token_positions_per_second": 9375.47862941137,
+      "required_pages_per_second_upper_estimate": 0.003925317119877616
     },
     {
       "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": 19111.47610109472,
-      "required_pages_per_second_upper_estimate": 0.003980416468340475
+      "required_token_positions_per_second": 18750.95725882274,
+      "required_pages_per_second_upper_estimate": 0.003905329901016446
     },
     {
       "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": 28667.214151642078,
-      "required_pages_per_second_upper_estimate": 0.0017185557160894315
+      "required_token_positions_per_second": 28126.435888234104,
+      "required_pages_per_second_upper_estimate": 0.001686136884918721
     },
     {
       "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": 28667.214151642078,
-      "required_pages_per_second_upper_estimate": 0.002228749806541117
+      "required_token_positions_per_second": 28126.435888234104,
+      "required_pages_per_second_upper_estimate": 0.0021867066751933457
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28667.214151642078,
-      "required_pages_per_second_upper_estimate": 0.36771224268085423
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28126.435888234104,
+      "required_pages_per_second_upper_estimate": 0.3607757197603174
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,42 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10924110078121885,
-    "producer_p95_seconds": 0.32970197800023016,
-    "gpu_update_seconds": 0.9644466969740868,
-    "producer_batches_per_second": 9.154063743853484,
-    "gpu_batches_per_second": 1.0368639377764062,
-    "producer_headroom": 8.828606541649735,
+    "producer_batch_seconds": 0.12665232875011156,
+    "producer_p95_seconds": 0.3157519520027563,
+    "gpu_update_seconds": 0.9829898146307885,
+    "producer_batches_per_second": 7.895630580729604,
+    "gpu_batches_per_second": 1.0173045387816155,
+    "producer_headroom": 7.761324440944578,
     "sustainable": true,
-    "buffer_seconds": 60.76014190936747,
+    "buffer_seconds": 60.94536850710889,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 340180.5052731563,
-    "upload_bytes_per_second": 5134244.0499124015,
+    "generation_bytes_per_second": 339143.57059869746,
+    "upload_bytes_per_second": 4912328.761082806,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.99481743474976,
+    "estimated_full_shards_per_hour": 72.77231539221471,
     "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": 1.1959222149998823,
-      "bytes": 7412774,
-      "pages": 30,
-      "shards": 3
-    },
-    {
-      "seconds": 1.488441301000421,
-      "bytes": 6011041,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0648157450013969,
       "bytes": 668578,
@@ -562,6 +550,18 @@
       "bytes": 12461215,
       "pages": 46,
       "shards": 4
+    },
+    {
+      "seconds": 1.7147560890007298,
+      "bytes": 951064,
+      "pages": 27,
+      "shards": 1
+    },
+    {
+      "seconds": 2.651306643998396,
+      "bytes": 1793185,
+      "pages": 11,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -298,3 +298,4 @@
 {"step": 954800, "loss": 0.5444334745407104, "lr": 3e-05, "grad_norm": 0.35233121796476646, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139756502665, "tokens_per_second": 150143.417514215, "data_wait_s": 1.5052999515319243e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 954900, "loss": 0.5239086151123047, "lr": 3e-05, "grad_norm": 0.34884229323315963, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139770930518, "tokens_per_second": 149787.61962677827, "data_wait_s": 1.3266002497402951e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 955000, "loss": 0.5055690407752991, "lr": 3e-05, "grad_norm": 0.33062625529615614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139785457962, "tokens_per_second": 150277.44072599962, "data_wait_s": 1.7883001419249922e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 955100, "loss": 0.5422380566596985, "lr": 3e-05, "grad_norm": 0.3728766214485024, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139800204381, "tokens_per_second": 137296.19085368796, "data_wait_s": 1.4479999663308263e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f9416a47024c6e8f3590cada4af202d65f04d98536cdd7f8d5a5a89c737fb659
-size 640263
+oid sha256:bc0ca230bde10cb8c0eb2132f61591a1db17ab2e4b51b9cc96eff7ad53e5148a
+size 642381