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": 75390.84150036154,
-      "required_pages_per_second_upper_estimate": 0.0052157133458480606
+      "required_token_positions_per_second": 76823.04236404208,
+      "required_pages_per_second_upper_estimate": 0.0053147963247613264
     },
     {
       "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": 9423.855187545192,
-      "required_pages_per_second_upper_estimate": 0.0039455713745508375
+      "required_token_positions_per_second": 9602.88029550526,
+      "required_pages_per_second_upper_estimate": 0.004020525448784349
     },
     {
       "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": 18847.710375090384,
-      "required_pages_per_second_upper_estimate": 0.003925481023583736
+      "required_token_positions_per_second": 19205.76059101052,
+      "required_pages_per_second_upper_estimate": 0.0040000534411407315
     },
     {
       "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": 28271.565562635573,
-      "required_pages_per_second_upper_estimate": 0.0016948371872989073
+      "required_token_positions_per_second": 28808.640886515775,
+      "required_pages_per_second_upper_estimate": 0.0017270340328989892
     },
     {
       "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": 28271.565562635573,
-      "required_pages_per_second_upper_estimate": 0.002197989869020086
+      "required_token_positions_per_second": 28808.640886515775,
+      "required_pages_per_second_upper_estimate": 0.0022397451131070823
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28271.565562635573,
-      "required_pages_per_second_upper_estimate": 0.3626372873954358
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28808.640886515775,
+      "required_pages_per_second_upper_estimate": 0.36952631298361716
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.10017849978905247,
-    "producer_p95_seconds": 0.22524926999903983,
-    "gpu_update_seconds": 0.9779437201221112,
-    "producer_batches_per_second": 9.98218182649687,
-    "gpu_batches_per_second": 1.0225537312874557,
-    "producer_headroom": 9.76201203033968,
+    "producer_batch_seconds": 0.06095308956247436,
+    "producer_p95_seconds": 0.19940804899852083,
+    "gpu_update_seconds": 0.9597120568412851,
+    "producer_batches_per_second": 16.406059269153896,
+    "gpu_batches_per_second": 1.0419791987310394,
+    "producer_headroom": 15.745092885859716,
     "sustainable": true,
-    "buffer_seconds": 60.63251064757089,
+    "buffer_seconds": 60.461859581000965,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 156,
+    "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 600719.1770247575,
-    "upload_bytes_per_second": 5390797.625399134,
+    "generation_bytes_per_second": 610576.5550299477,
+    "upload_bytes_per_second": 5350218.939287733,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.4501756813862,
+    "estimated_full_shards_per_hour": 65.5077576073352,
     "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": 1.2558460740001465,
-      "bytes": 9834347,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.1021917439998106,
       "bytes": 2521958,
@@ -562,6 +556,12 @@
       "bytes": 1156490,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 1.3950513529998716,
+      "bytes": 7312918,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -106,3 +106,4 @@
 {"step": 895600, "loss": 0.5518725514411926, "lr": 3e-05, "grad_norm": 0.3277005091052649, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131090877456, "tokens_per_second": 153869.18445872568, "data_wait_s": 1.3012999261263758e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 895700, "loss": 0.5045432448387146, "lr": 3e-05, "grad_norm": 0.3197829840439839, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131105489841, "tokens_per_second": 151838.63309223627, "data_wait_s": 1.6257999959634617e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 895800, "loss": 0.5227249264717102, "lr": 3e-05, "grad_norm": 0.3118212020557827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131119945106, "tokens_per_second": 151109.3078300012, "data_wait_s": 1.5307001376640983e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 895900, "loss": 0.52464759349823, "lr": 3e-05, "grad_norm": 0.3165974995314781, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 131134480203, "tokens_per_second": 152016.57642530568, "data_wait_s": 1.2111999240005389e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c3534a8ceeeb1b00cc4608d56c51897201210197e51aa6b1542fc5b9e128fd88
-size 230475
+oid sha256:df69943235d7ab6c7bf1890546146cfcfe636c289956d9e38af88fcc11779b0b
+size 232593