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": 75703.29419945383,
-      "required_pages_per_second_upper_estimate": 0.006561769672963654
+      "required_token_positions_per_second": 76173.4388520347,
+      "required_pages_per_second_upper_estimate": 0.006602520619878636
     },
     {
       "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": 9462.911774931728,
-      "required_pages_per_second_upper_estimate": 0.0037748969901594577
+      "required_token_positions_per_second": 9521.679856504337,
+      "required_pages_per_second_upper_estimate": 0.003798340456559892
     },
     {
       "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": 18925.823549863457,
-      "required_pages_per_second_upper_estimate": 0.004378156195142634
+      "required_token_positions_per_second": 19043.359713008675,
+      "required_pages_per_second_upper_estimate": 0.004405346117920459
     },
     {
       "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": 28388.735324795183,
-      "required_pages_per_second_upper_estimate": 0.0017287250370967904
+      "required_token_positions_per_second": 28565.03956951301,
+      "required_pages_per_second_upper_estimate": 0.0017394610406032216
     },
     {
       "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": 28388.735324795183,
-      "required_pages_per_second_upper_estimate": 0.002185443430351779
+      "required_token_positions_per_second": 28565.03956951301,
+      "required_pages_per_second_upper_estimate": 0.002199015819151543
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28388.735324795183,
-      "required_pages_per_second_upper_estimate": 0.35249308174870164
+      "required_token_positions_per_second": 28565.03956951301,
+      "required_pages_per_second_upper_estimate": 0.35468219041574695
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.2189259103746508,
-    "producer_p95_seconds": 0.37597236600413453,
-    "gpu_update_seconds": 0.9739074208019329,
-    "producer_batches_per_second": 4.567755357457172,
-    "gpu_batches_per_second": 1.0267916422451961,
-    "producer_headroom": 4.448570839035326,
+    "producer_batch_seconds": 0.12146320373415165,
+    "producer_p95_seconds": 0.38165004400070757,
+    "gpu_update_seconds": 0.967896436226479,
+    "producer_batches_per_second": 8.232946021979751,
+    "gpu_batches_per_second": 1.033168387207502,
+    "producer_headroom": 7.96863911431917,
     "sustainable": true,
-    "buffer_seconds": 60.38226008971984,
+    "buffer_seconds": 60.0095790460417,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 375860.6840113863,
-    "upload_bytes_per_second": 5081827.115893066,
+    "generation_bytes_per_second": 371972.12746038596,
+    "upload_bytes_per_second": 5141309.849263821,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.65095319992248,
+    "estimated_full_shards_per_hour": 79.81655948504147,
     "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": 1.3213498320001236,
-      "bytes": 5953362,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9344346500001848,
       "bytes": 5832415,
@@ -562,6 +556,12 @@
       "bytes": 6082599,
       "pages": 34,
       "shards": 2
+    },
+    {
+      "seconds": 1.4744428199992399,
+      "bytes": 11695751,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -378,3 +378,4 @@
 {"step": 192800, "loss": 0.6202797312289476, "lr": 0.0003, "grad_norm": 0.15110603796866887, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28218390437, "tokens_per_second": 150834.49217930407, "data_wait_s": 9.115006832871586e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 192900, "loss": 0.6916948650032282, "lr": 0.0003, "grad_norm": 0.1605090625193179, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28233021763, "tokens_per_second": 149785.7353450285, "data_wait_s": 1.3833996490575373e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 193000, "loss": 0.5843979194760323, "lr": 0.0003, "grad_norm": 0.151001145027427, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28247476705, "tokens_per_second": 148024.0681398775, "data_wait_s": 1.2589000107254833e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 193100, "loss": 0.5864177253097296, "lr": 0.0003, "grad_norm": 0.1442607025269688, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28262060521, "tokens_per_second": 150170.15379304672, "data_wait_s": 1.2289005098864436e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8a71d823498fb7e636525b1f22ef0fdda8dc3a8ea9f484d8212633c853628ab0
-size 810486
+oid sha256:5bd3b686c82800763b256366e7770e47937a4a06ffbf657fc5e87546530d0fb9
+size 812604