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": 77187.22500558228,
-      "required_pages_per_second_upper_estimate": 0.005339991324923227
+      "required_token_positions_per_second": 75833.23133098953,
+      "required_pages_per_second_upper_estimate": 0.0052463189008167335
     },
     {
       "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": 9648.403125697785,
-      "required_pages_per_second_upper_estimate": 0.004039584907161165
+      "required_token_positions_per_second": 9479.15391637369,
+      "required_pages_per_second_upper_estimate": 0.003968723797542559
     },
     {
       "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": 19296.80625139557,
-      "required_pages_per_second_upper_estimate": 0.0040190158511634105
+      "required_token_positions_per_second": 18958.30783274738,
+      "required_pages_per_second_upper_estimate": 0.003948515557362594
     },
     {
       "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": 28945.209377093353,
-      "required_pages_per_second_upper_estimate": 0.0017352211053811041
+      "required_token_positions_per_second": 28437.461749121074,
+      "required_pages_per_second_upper_estimate": 0.0017047824103699034
     },
     {
       "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": 28945.209377093353,
-      "required_pages_per_second_upper_estimate": 0.002250362712548183
+      "required_token_positions_per_second": 28437.461749121074,
+      "required_pages_per_second_upper_estimate": 0.0022108875678191267
     },
     {
       "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": 28945.209377093353,
-      "required_pages_per_second_upper_estimate": 0.3712780669449257
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28437.461749121074,
+      "required_pages_per_second_upper_estimate": 0.3647652255502248
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +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": 0.11966711167167432,
-    "producer_p95_seconds": 0.3300721020059427,
-    "gpu_update_seconds": 0.9551839698171284,
-    "producer_batches_per_second": 8.356514885590775,
-    "gpu_batches_per_second": 1.0469187419376937,
-    "producer_headroom": 7.982009062254523,
+    "producer_batch_seconds": 0.11432970593006075,
+    "producer_p95_seconds": 0.34066012399853207,
+    "gpu_update_seconds": 0.9722386703818433,
+    "producer_batches_per_second": 8.746633185707072,
+    "gpu_batches_per_second": 1.0285540273842981,
+    "producer_headroom": 8.50381501878955,
     "sustainable": true,
-    "buffer_seconds": 60.17659009847909,
+    "buffer_seconds": 60.27879756367429,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 157,
+    "upload_pages_per_commit": 156,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 563357.6609029752,
-    "upload_bytes_per_second": 5376159.544282721,
+    "generation_bytes_per_second": 565652.8129276019,
+    "upload_bytes_per_second": 5295376.215007366,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.441719867310255,
+    "estimated_full_shards_per_hour": 60.68796296535035,
     "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": 1.0333158499997808,
-      "bytes": 5705052,
-      "pages": 24,
-      "shards": 2
-    },
     {
       "seconds": 1.4537645320015145,
       "bytes": 12446484,
@@ -562,6 +556,12 @@
       "bytes": 17764299,
       "pages": 34,
       "shards": 5
+    },
+    {
+      "seconds": 1.4270490490016527,
+      "bytes": 1152119,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -365,3 +365,4 @@
 {"step": 681500, "loss": 0.5778361558914185, "lr": 0.0001, "grad_norm": 0.19999356571613733, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99753818140, "tokens_per_second": 149664.32069253526, "data_wait_s": 1.180300023406744e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 681600, "loss": 0.5432363152503967, "lr": 0.0001, "grad_norm": 0.20279183690871785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99768308648, "tokens_per_second": 151271.44525978353, "data_wait_s": 1.5789999451953918e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 681700, "loss": 0.5722970366477966, "lr": 0.0001, "grad_norm": 0.19341553347340884, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99782992451, "tokens_per_second": 152805.225464494, "data_wait_s": 1.0833995474968106e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 681800, "loss": 0.6396709084510803, "lr": 0.0001, "grad_norm": 0.22740485523095524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99797753860, "tokens_per_second": 154039.79191743594, "data_wait_s": 1.2878997949883342e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1c80785399518aa6f166696221d14bbcd351d08929d33d2e78764fa1b3247ad2
-size 782952
+oid sha256:7129c85c0ec548df8db4f8d9e5dbab2010639f774c1c0a11146b855b7851591b
+size 785070