Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75791.46970273064,
-      "required_pages_per_second_upper_estimate": 0.006569412502108918
+      "required_token_positions_per_second": 74711.89297351196,
+      "required_pages_per_second_upper_estimate": 0.006475837527382448
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9473.93371284133,
-      "required_pages_per_second_upper_estimate": 0.003779293805984255
+      "required_token_positions_per_second": 9338.986621688995,
+      "required_pages_per_second_upper_estimate": 0.0037254613936847755
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18947.86742568266,
-      "required_pages_per_second_upper_estimate": 0.004383255657854443
+      "required_token_positions_per_second": 18677.97324337799,
+      "required_pages_per_second_upper_estimate": 0.004320820388753634
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28421.80113852399,
-      "required_pages_per_second_upper_estimate": 0.0017307385716699578
+      "required_token_positions_per_second": 28016.959865066983,
+      "required_pages_per_second_upper_estimate": 0.0017060858621544333
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28421.80113852399,
-      "required_pages_per_second_upper_estimate": 0.0021879889282246535
+      "required_token_positions_per_second": 28016.959865066983,
+      "required_pages_per_second_upper_estimate": 0.0021568231263215617
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28421.80113852399,
-      "required_pages_per_second_upper_estimate": 0.3529036484910537
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28016.959865066983,
+      "required_pages_per_second_upper_estimate": 0.3478768747913007
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.11377870719530847,
-    "producer_p95_seconds": 0.2347504770004889,
-    "gpu_update_seconds": 0.972774380668115,
-    "producer_batches_per_second": 8.788990705294582,
-    "gpu_batches_per_second": 1.0279875990496234,
-    "producer_headroom": 8.549704990040757,
+    "producer_batch_seconds": 0.0681238520469094,
+    "producer_p95_seconds": 0.22342332699918188,
+    "gpu_update_seconds": 0.9868308386475928,
+    "producer_batches_per_second": 14.679146436279176,
+    "gpu_batches_per_second": 1.0133449025270176,
+    "producer_headroom": 14.485834388344202,
     "sustainable": true,
-    "buffer_seconds": 60.31201160142313,
+    "buffer_seconds": 60.196681157503164,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 427411.21125998616,
+    "generation_bytes_per_second": 427095.9085925818,
     "upload_bytes_per_second": 5770565.508757728,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 91.71249631261529,
+    "estimated_full_shards_per_hour": 91.64483970006076,
     "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": [

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -191,3 +191,4 @@
 {"step": 374100, "loss": 0.6167305372655392, "lr": 0.0003, "grad_norm": 0.15267573844008275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54757211934, "tokens_per_second": 149487.78632816183, "data_wait_s": 8.283997885882854e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 374200, "loss": 0.6089011132717133, "lr": 0.0003, "grad_norm": 0.17899571113931081, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54771927955, "tokens_per_second": 149511.47149465152, "data_wait_s": 1.4418001228477806e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 374300, "loss": 0.6218976192176342, "lr": 0.0003, "grad_norm": 0.14650837527254615, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54786729345, "tokens_per_second": 150609.22065152868, "data_wait_s": 7.520000508520752e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 374400, "loss": 0.5192107986658812, "lr": 0.0003, "grad_norm": 0.15734461036322073, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 54801338518, "tokens_per_second": 147650.84406482143, "data_wait_s": 1.2248001439729705e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c530a7865d512d280482ad96a12ef6ba05d4e145bad600d69d61e855901b3d1a
-size 411288
+oid sha256:3cabbb3d0a3320c24444195c5d5a8fc960c915be6b1940b8c997c92a4af40246
+size 413406