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": 75388.70894284581,
-      "required_pages_per_second_upper_estimate": 0.006534502220229915
+      "required_token_positions_per_second": 74467.55649924783,
+      "required_pages_per_second_upper_estimate": 0.006454659061058366
     },
     {
       "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": 9423.588617855727,
-      "required_pages_per_second_upper_estimate": 0.003759210394868249
+      "required_token_positions_per_second": 9308.444562405979,
+      "required_pages_per_second_upper_estimate": 0.0037132777095923004
     },
     {
       "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": 18847.177235711453,
-      "required_pages_per_second_upper_estimate": 0.0043599627544914235
+      "required_token_positions_per_second": 18616.889124811958,
+      "required_pages_per_second_upper_estimate": 0.004306689652967152
     },
     {
       "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": 28270.765853567184,
-      "required_pages_per_second_upper_estimate": 0.0017215413152369867
+      "required_token_positions_per_second": 27925.333687217935,
+      "required_pages_per_second_upper_estimate": 0.0017005063086488292
     },
     {
       "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": 28270.765853567184,
-      "required_pages_per_second_upper_estimate": 0.0021763618138962507
+      "required_token_positions_per_second": 27925.333687217935,
+      "required_pages_per_second_upper_estimate": 0.0021497694895132482
     },
     {
       "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": 28270.765853567184,
-      "required_pages_per_second_upper_estimate": 0.35102829573447214
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27925.333687217935,
+      "required_pages_per_second_upper_estimate": 0.3467391843155064
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.12732536664026384,
-    "producer_p95_seconds": 0.41907933500624495,
-    "gpu_update_seconds": 0.97797138369746,
-    "producer_batches_per_second": 7.853894525395948,
-    "gpu_batches_per_second": 1.0225248066249704,
-    "producer_headroom": 7.68088409641538,
+    "producer_batch_seconds": 0.13499817716416374,
+    "producer_p95_seconds": 0.44701500100200064,
+    "gpu_update_seconds": 0.990068742227962,
+    "producer_batches_per_second": 7.407507427185153,
+    "gpu_batches_per_second": 1.0100308769971764,
+    "producer_headroom": 7.333941561477491,
     "sustainable": true,
-    "buffer_seconds": 60.634225789242514,
+    "buffer_seconds": 60.394193275905685,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 498548.53900297923,
-    "upload_bytes_per_second": 4676279.879999559,
+    "generation_bytes_per_second": 478133.5785813136,
+    "upload_bytes_per_second": 4628037.193657025,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 106.9769108540252,
+    "estimated_full_shards_per_hour": 102.59633558349186,
     "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": 1.4798621720037772,
-      "bytes": 13649971,
-      "pages": 37,
-      "shards": 4
-    },
     {
       "seconds": 0.8202600639997399,
       "bytes": 1256174,
@@ -562,6 +556,12 @@
       "bytes": 5487074,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.1849965610017534,
+      "bytes": 7913091,
+      "pages": 40,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -411,3 +411,4 @@
 {"step": 196100, "loss": 0.5884687602519989, "lr": 0.0003, "grad_norm": 0.13519273574053875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28701200508, "tokens_per_second": 146497.09774357386, "data_wait_s": 9.642004442866892e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 196200, "loss": 0.5501316413283348, "lr": 0.0003, "grad_norm": 0.12685191393783454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28715614316, "tokens_per_second": 147621.3208553125, "data_wait_s": 1.270500069949776e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 196300, "loss": 0.6403248496353626, "lr": 0.0003, "grad_norm": 0.17848715324458939, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28730226039, "tokens_per_second": 148628.79923139638, "data_wait_s": 1.3712997315451503e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 196400, "loss": 0.6060383096337318, "lr": 0.0003, "grad_norm": 0.15123548976325515, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 28744944508, "tokens_per_second": 150985.67238656757, "data_wait_s": 1.1547999747563154e-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 sha256:9844c0d41c9f61f4a4d4a8db4c80949f01e4b5447420f5b5a75018b2ce44ca1a
-size 881163
+oid sha256:6017549657a8476e503cd4550543ec6b00ed1f47eb0879a6e961273981f54445
+size 883281