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": 75423.77351164505,
-      "required_pages_per_second_upper_estimate": 0.005217991658801447
+      "required_token_positions_per_second": 76307.95135286232,
+      "required_pages_per_second_upper_estimate": 0.005279161133432101
     },
     {
       "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": 9427.971688955631,
-      "required_pages_per_second_upper_estimate": 0.0039472948677290644
+      "required_token_positions_per_second": 9538.49391910779,
+      "required_pages_per_second_upper_estimate": 0.00399356821752716
     },
     {
       "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": 18855.943377911262,
-      "required_pages_per_second_upper_estimate": 0.00392719574095244
+      "required_token_positions_per_second": 19076.98783821558,
+      "required_pages_per_second_upper_estimate": 0.003973233472699405
     },
     {
       "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": 28283.91506686689,
-      "required_pages_per_second_upper_estimate": 0.0016955775212209737
+      "required_token_positions_per_second": 28615.481757323367,
+      "required_pages_per_second_upper_estimate": 0.001715454438040817
     },
     {
       "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": 28283.91506686689,
-      "required_pages_per_second_upper_estimate": 0.002198949988654343
+      "required_token_positions_per_second": 28615.481757323367,
+      "required_pages_per_second_upper_estimate": 0.0022247278404295915
     },
     {
       "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": 28283.91506686689,
-      "required_pages_per_second_upper_estimate": 0.3627956935758506
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28615.481757323367,
+      "required_pages_per_second_upper_estimate": 0.3670486750724512
     }
   ],
   "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": 0.47393304695305005,
-    "producer_p95_seconds": 0.34209822199773043,
-    "gpu_update_seconds": 0.9775167240686622,
-    "producer_batches_per_second": 2.110002681663734,
-    "gpu_batches_per_second": 1.0230004002773037,
-    "producer_headroom": 2.062562909156026,
+    "producer_batch_seconds": 0.0954549049924367,
+    "producer_p95_seconds": 0.32441666300292127,
+    "gpu_update_seconds": 0.9661902684173483,
+    "producer_batches_per_second": 10.47615101685172,
+    "gpu_batches_per_second": 1.0349928297643,
+    "producer_headroom": 10.12195516295264,
     "sustainable": true,
-    "buffer_seconds": 60.60603689225706,
+    "buffer_seconds": 60.869986910292944,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 85,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 244485.8707781613,
-    "upload_bytes_per_second": 3635378.9169689966,
+    "generation_bytes_per_second": 244150.69501919032,
+    "upload_bytes_per_second": 3666832.9400221235,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 52.4609765292037,
+    "estimated_full_shards_per_hour": 52.38905561382086,
     "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": 1.5062627519982925,
-      "bytes": 6228251,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.0129257990010956,
       "bytes": 2288801,
@@ -562,6 +556,12 @@
       "bytes": 859687,
       "pages": 24,
       "shards": 1
+    },
+    {
+      "seconds": 1.6816289050038904,
+      "bytes": 10458494,
+      "pages": 47,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -383,3 +383,4 @@
 {"step": 880800, "loss": 0.5035049915313721, "lr": 3e-05, "grad_norm": 0.29853296905549626, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128924604752, "tokens_per_second": 153178.8784803436, "data_wait_s": 1.2065000191796571e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 880900, "loss": 0.5186980366706848, "lr": 3e-05, "grad_norm": 0.32294184104601764, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128939072513, "tokens_per_second": 149594.84237675718, "data_wait_s": 1.2589000107254833e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 881000, "loss": 0.5027999877929688, "lr": 3e-05, "grad_norm": 0.29615730230163156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128953639377, "tokens_per_second": 150685.19613681379, "data_wait_s": 1.4551995263900608e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 881100, "loss": 0.5287010073661804, "lr": 3e-05, "grad_norm": 0.3909493589346569, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 128968269669, "tokens_per_second": 151252.52617220164, "data_wait_s": 1.4672994439024478e-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 sha256:524ac477d281b507f547d3cd7c5ed33e7e8e7238b4e3351042029db0612a8a59
-size 821859
+oid sha256:0c0fb2ab7d0dec679be0d26dd1624b58439fffbb44a09763cd4952b432e36abc
+size 823977