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": 73638.15505737084,
-      "required_pages_per_second_upper_estimate": 0.005094458430929121
+      "required_token_positions_per_second": 77141.36184015001,
+      "required_pages_per_second_upper_estimate": 0.005336818404721418
     },
     {
       "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": 9204.769382171355,
-      "required_pages_per_second_upper_estimate": 0.0038538447228726722
+      "required_token_positions_per_second": 9642.670230018752,
+      "required_pages_per_second_upper_estimate": 0.004037184663456829
     },
     {
       "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": 18409.53876434271,
-      "required_pages_per_second_upper_estimate": 0.003834221432427435
+      "required_token_positions_per_second": 19285.340460037503,
+      "required_pages_per_second_upper_estimate": 0.004016627829196776
     },
     {
       "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": 27614.308146514064,
-      "required_pages_per_second_upper_estimate": 0.0016554356087763883
+      "required_token_positions_per_second": 28928.010690056253,
+      "required_pages_per_second_upper_estimate": 0.0017341900703541045
     },
     {
       "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": 27614.308146514064,
-      "required_pages_per_second_upper_estimate": 0.002146890995886485
+      "required_token_positions_per_second": 28928.010690056253,
+      "required_pages_per_second_upper_estimate": 0.002249025590280078
     },
     {
       "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": 27614.308146514064,
-      "required_pages_per_second_upper_estimate": 0.3542066949694599
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28928.010690056253,
+      "required_pages_per_second_upper_estimate": 0.3710574606541252
     }
   ],
   "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": 60,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.23797507191409295,
-    "producer_p95_seconds": 0.23150020399771165,
-    "gpu_update_seconds": 1.001220086822642,
-    "producer_batches_per_second": 4.202120801799744,
-    "gpu_batches_per_second": 0.9987813999751903,
-    "producer_headroom": 4.207247754017169,
+    "producer_batch_seconds": 0.2928047587032552,
+    "producer_p95_seconds": 0.23311134800314903,
+    "gpu_update_seconds": 0.9557518592007349,
+    "producer_batches_per_second": 3.4152450405133488,
+    "gpu_batches_per_second": 1.046296682944743,
+    "producer_headroom": 3.264126797096722,
     "sustainable": true,
-    "buffer_seconds": 60.07320520935852,
+    "buffer_seconds": 60.212367129646296,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 339148.33964203927,
-    "upload_bytes_per_second": 3477540.94489272,
+    "generation_bytes_per_second": 366219.1737603757,
+    "upload_bytes_per_second": 3548393.8583439505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.77333871789821,
+    "estimated_full_shards_per_hour": 78.5821095429273,
     "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": 1.7027444539999124,
-      "bytes": 6185388,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.5859740610012523,
       "bytes": 2563401,
@@ -562,6 +556,12 @@
       "bytes": 1066838,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.2431204950007668,
+      "bytes": 13191953,
+      "pages": 36,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -183,3 +183,4 @@
 {"step": 860800, "loss": 0.5383110046386719, "lr": 3e-05, "grad_norm": 0.33065496086550183, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125996441007, "tokens_per_second": 150179.68485327312, "data_wait_s": 1.641799826757051e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 860900, "loss": 0.5218645334243774, "lr": 3e-05, "grad_norm": 0.3049896695397522, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126010991566, "tokens_per_second": 151597.86125618493, "data_wait_s": 1.4173998351907358e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 861000, "loss": 0.5609935522079468, "lr": 3e-05, "grad_norm": 0.29852792585960003, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126025698838, "tokens_per_second": 151450.16261014933, "data_wait_s": 1.896000321721658e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 861100, "loss": 0.6365047693252563, "lr": 3e-05, "grad_norm": 0.3565185435318611, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 126040404087, "tokens_per_second": 152822.69922314503, "data_wait_s": 1.1434000043664128e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9ccca67e883c93f30057e54073ae89f5288796d390779d96943a89e5900028b0
-size 395127
+oid sha256:36e94df220ffda130622c74ba7adb6d73ec6a817127d70645542fb01a14187cf
+size 397245