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": 76751.80172971137,
-      "required_pages_per_second_upper_estimate": 0.005309867732377282
+      "required_token_positions_per_second": 76697.08149627564,
+      "required_pages_per_second_upper_estimate": 0.005306082059659764
     },
     {
       "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": 9593.975216213921,
-      "required_pages_per_second_upper_estimate": 0.004016797078044266
+      "required_token_positions_per_second": 9587.135187034455,
+      "required_pages_per_second_upper_estimate": 0.004013933300662876
     },
     {
       "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": 19187.950432427842,
-      "required_pages_per_second_upper_estimate": 0.0039963440547934265
+      "required_token_positions_per_second": 19174.27037406891,
+      "required_pages_per_second_upper_estimate": 0.0039934948594046605
     },
     {
       "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": 28781.925648641765,
-      "required_pages_per_second_upper_estimate": 0.0017254324951802486
+      "required_token_positions_per_second": 28761.405561103365,
+      "required_pages_per_second_upper_estimate": 0.001724202347264681
     },
     {
       "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": 28781.925648641765,
-      "required_pages_per_second_upper_estimate": 0.0022376681208702907
+      "required_token_positions_per_second": 28761.405561103365,
+      "required_pages_per_second_upper_estimate": 0.0022360727743225067
     },
     {
       "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": 28781.925648641765,
-      "required_pages_per_second_upper_estimate": 0.3691836385967569
+      "required_token_positions_per_second": 28761.405561103365,
+      "required_pages_per_second_upper_estimate": 0.36892042894656774
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05217598170312954,
-    "producer_p95_seconds": 0.17138480100038578,
-    "gpu_update_seconds": 0.9606028567204198,
-    "producer_batches_per_second": 19.165906751688766,
-    "gpu_batches_per_second": 1.0410129357871007,
-    "producer_headroom": 18.41082477730941,
+    "producer_batch_seconds": 0.06993126337498268,
+    "producer_p95_seconds": 0.1820425140003863,
+    "gpu_update_seconds": 0.961288207603834,
+    "producer_batches_per_second": 14.299755956614701,
+    "gpu_batches_per_second": 1.040270745120926,
+    "producer_headroom": 13.746186772706395,
     "sustainable": true,
-    "buffer_seconds": 60.517979973386446,
+    "buffer_seconds": 60.56115707904154,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 345024.7708426248,
-    "upload_bytes_per_second": 4955157.3166293865,
+    "generation_bytes_per_second": 358200.85358049226,
+    "upload_bytes_per_second": 4850326.766404005,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 74.03428405722674,
+    "estimated_full_shards_per_hour": 76.86156468926502,
     "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": [
@@ -346,6 +346,12 @@
       "bytes": 17752294,
       "pages": 35,
       "shards": 5
+    },
+    {
+      "seconds": 1.054843687000357,
+      "bytes": 1282245,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -73,3 +73,4 @@
 {"step": 727300, "loss": 0.5268535017967224, "lr": 5e-05, "grad_norm": 0.23538101980784507, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106456330822, "tokens_per_second": 129230.42199775409, "data_wait_s": 1.3830000170855783e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 727400, "loss": 0.5175501108169556, "lr": 5e-05, "grad_norm": 0.22082917564890717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106470834723, "tokens_per_second": 144143.5542440665, "data_wait_s": 1.110399989556754e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 727500, "loss": 0.5392053127288818, "lr": 5e-05, "grad_norm": 0.2323101730866226, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106485432700, "tokens_per_second": 150974.56391772357, "data_wait_s": 1.2364999747660477e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 727600, "loss": 0.561463475227356, "lr": 5e-05, "grad_norm": 0.21850676530010643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106500106885, "tokens_per_second": 152043.26874381435, "data_wait_s": 1.2847999641962815e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:441734652af083998e55ae6c2a5fe81c96375ff8f9ee59909db8e35ee9ab5b01
-size 159798
+oid sha256:10fc46a3b88964ad3f2deefd8487a2b1a68827f8f62fc577b79f538145c70fbc
+size 161916