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": 75152.664799365,
-      "required_pages_per_second_upper_estimate": 0.00651404251212141
+      "required_token_positions_per_second": 74535.87286666177,
+      "required_pages_per_second_upper_estimate": 0.006460580550639555
     },
     {
       "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": 9394.083099920625,
-      "required_pages_per_second_upper_estimate": 0.0037474402026171313
+      "required_token_positions_per_second": 9316.984108332721,
+      "required_pages_per_second_upper_estimate": 0.0037166842621400675
     },
     {
       "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": 18788.16619984125,
-      "required_pages_per_second_upper_estimate": 0.004346311589901612
+      "required_token_positions_per_second": 18633.968216665442,
+      "required_pages_per_second_upper_estimate": 0.004310640600285705
     },
     {
       "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": 28182.249299761876,
-      "required_pages_per_second_upper_estimate": 0.0017161511215207097
+      "required_token_positions_per_second": 27950.95232499816,
+      "required_pages_per_second_upper_estimate": 0.0017020663492790403
     },
     {
       "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": 28182.249299761876,
-      "required_pages_per_second_upper_estimate": 0.0021695475645548143
+      "required_token_positions_per_second": 27950.95232499816,
+      "required_pages_per_second_upper_estimate": 0.0021517416831664986
     },
     {
       "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": 28182.249299761876,
-      "required_pages_per_second_upper_estimate": 0.34992921638205887
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27950.95232499816,
+      "required_pages_per_second_upper_estimate": 0.34705728205667163
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.0593059881874467,
-    "producer_p95_seconds": 0.20257730599951174,
-    "gpu_update_seconds": 0.9810430567809082,
-    "producer_batches_per_second": 16.86170369237132,
-    "gpu_batches_per_second": 1.0193232530295817,
-    "producer_headroom": 16.542057332897887,
+    "producer_batch_seconds": 0.3424345556173165,
+    "producer_p95_seconds": 0.22388275299999805,
+    "gpu_update_seconds": 0.9891612879062008,
+    "producer_batches_per_second": 2.9202660292191354,
+    "gpu_batches_per_second": 1.0109574770326302,
+    "producer_headroom": 2.888614106491127,
     "sustainable": true,
-    "buffer_seconds": 60.82466952041631,
+    "buffer_seconds": 60.33883856227825,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 409560.69388242485,
+    "generation_bytes_per_second": 395944.2546959774,
     "upload_bytes_per_second": 5868308.488682985,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 87.88219082216796,
+    "estimated_full_shards_per_hour": 84.96041994723788,
     "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-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -86,3 +86,4 @@
 {"step": 123600, "loss": 0.645138205960393, "lr": 0.0003, "grad_norm": 0.13218399506101208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18087558450, "tokens_per_second": 148748.44300659816, "data_wait_s": 1.629899998079054e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 123700, "loss": 0.6238192729651928, "lr": 0.0003, "grad_norm": 0.13839264763992487, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18102136122, "tokens_per_second": 148817.6636587398, "data_wait_s": 8.29599957796745e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 123800, "loss": 0.5984497033059597, "lr": 0.0003, "grad_norm": 0.1503882888211113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18116607536, "tokens_per_second": 147450.02039798637, "data_wait_s": 1.4983999790274538e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 123900, "loss": 0.6461109630763531, "lr": 0.0003, "grad_norm": 0.13312823231443432, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18131195088, "tokens_per_second": 148245.91900533746, "data_wait_s": 1.7831000150181353e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9bb892071f1c3591fae62de5d83c85f89c2005980ecce2ef5d97657393dcbfa0
-size 187332
+oid sha256:831e0af0fa316fbd58af70119a53e1556ddf252b4d88eb31b952d13b1d1be673
+size 189450