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": 75832.88862808807,
-      "required_pages_per_second_upper_estimate": 0.006573002589583593
+      "required_token_positions_per_second": 74731.54033366744,
+      "required_pages_per_second_upper_estimate": 0.0064775405107649455
     },
     {
       "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": 9479.111078511009,
-      "required_pages_per_second_upper_estimate": 0.0037813591345584045
+      "required_token_positions_per_second": 9341.44254170843,
+      "required_pages_per_second_upper_estimate": 0.003726441096899804
     },
     {
       "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": 18958.222157022017,
-      "required_pages_per_second_upper_estimate": 0.004385651042712756
+      "required_token_positions_per_second": 18682.88508341686,
+      "required_pages_per_second_upper_estimate": 0.004321956656501195
     },
     {
       "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": 28437.333235533024,
-      "required_pages_per_second_upper_estimate": 0.0017316843948871927
+      "required_token_positions_per_second": 28024.327625125286,
+      "required_pages_per_second_upper_estimate": 0.0017065345200863345
     },
     {
       "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": 28437.333235533024,
-      "required_pages_per_second_upper_estimate": 0.0021891846320480053
+      "required_token_positions_per_second": 28024.327625125286,
+      "required_pages_per_second_upper_estimate": 0.0021573903168861154
     },
     {
       "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": 28437.333235533024,
-      "required_pages_per_second_upper_estimate": 0.35309650515332114
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28024.327625125286,
+      "required_pages_per_second_upper_estimate": 0.3479683577129181
     }
   ],
   "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": 717798,
-    "producer_batch_seconds": 0.08194227548437993,
-    "producer_p95_seconds": 0.3199367960005475,
-    "gpu_update_seconds": 0.9722430641088828,
-    "producer_batches_per_second": 12.203712846497933,
-    "gpu_batches_per_second": 1.028549379178712,
-    "producer_headroom": 11.864975171384087,
+    "producer_batch_seconds": 0.2280991237732053,
+    "producer_p95_seconds": 0.3375095799965493,
+    "gpu_update_seconds": 0.9865713950336532,
+    "producer_batches_per_second": 4.384058927794397,
+    "gpu_batches_per_second": 1.0136113869041263,
+    "producer_headroom": 4.325187132303861,
     "sustainable": true,
-    "buffer_seconds": 60.27906997475073,
+    "buffer_seconds": 60.18085509705285,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 172,
+    "upload_pages_per_commit": 164,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 685715.9656808543,
-    "upload_bytes_per_second": 4847960.428993928,
+    "generation_bytes_per_second": 648898.2266022367,
+    "upload_bytes_per_second": 4906199.64481239,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.56934179219829,
+    "estimated_full_shards_per_hour": 69.61922692561305,
     "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": 1.2923507409977901,
-      "bytes": 7469632,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.187904915001127,
       "bytes": 6100274,
@@ -562,6 +556,12 @@
       "bytes": 2707013,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 0.9819577199996274,
+      "bytes": 10903211,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -273,3 +273,4 @@
 {"step": 382300, "loss": 0.6177261415868998, "lr": 0.0003, "grad_norm": 0.15430029791062644, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55957440697, "tokens_per_second": 148790.28635864134, "data_wait_s": 1.2992997653782368e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 382400, "loss": 0.5885235872119665, "lr": 0.0003, "grad_norm": 0.1462610040250813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55971996264, "tokens_per_second": 147671.78293035118, "data_wait_s": 1.0176001524087042e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 382500, "loss": 0.587795801460743, "lr": 0.0003, "grad_norm": 0.17074154889065904, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 55986664145, "tokens_per_second": 148913.49731703813, "data_wait_s": 1.3286000466905534e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 382600, "loss": 0.6028595305979252, "lr": 0.0003, "grad_norm": 0.17578412955566114, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56001388975, "tokens_per_second": 150112.0015920155, "data_wait_s": 1.0195999493589625e-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 sha256:578f68e04fc2526754f53a422af15ef45c8bf312dcd87957278c60221f747449
-size 586530
+oid sha256:432a425120450ada02f35e9b48c8589791d682ec94ae018df2b6564f63e2f58e
+size 588648