Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 77756.38875596202,
-      "required_pages_per_second_upper_estimate": 0.005379367393816347
+      "required_token_positions_per_second": 77225.44782083525,
+      "required_pages_per_second_upper_estimate": 0.005342635667971578
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9719.548594495252,
-      "required_pages_per_second_upper_estimate": 0.004069372029260333
+      "required_token_positions_per_second": 9653.180977604407,
+      "required_pages_per_second_upper_estimate": 0.004041585293981574
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19439.097188990505,
-      "required_pages_per_second_upper_estimate": 0.004048651301000068
+      "required_token_positions_per_second": 19306.361955208813,
+      "required_pages_per_second_upper_estimate": 0.004021006052266853
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29158.645783485754,
-      "required_pages_per_second_upper_estimate": 0.001748016292045806
+      "required_token_positions_per_second": 28959.54293281322,
+      "required_pages_per_second_upper_estimate": 0.0017360803801604377
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 29158.645783485754,
-      "required_pages_per_second_upper_estimate": 0.0022669564543376566
+      "required_token_positions_per_second": 28959.54293281322,
+      "required_pages_per_second_upper_estimate": 0.0022514770834587542
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29158.645783485754,
-      "required_pages_per_second_upper_estimate": 0.3740157999959692
+      "required_token_positions_per_second": 28959.54293281322,
+      "required_pages_per_second_upper_estimate": 0.37146192240752707
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.20816273722672918,
-    "producer_p95_seconds": 0.22999346700089518,
-    "gpu_update_seconds": 0.9481921830422824,
-    "producer_batches_per_second": 4.803933755496345,
-    "gpu_batches_per_second": 1.0546385193679744,
-    "producer_headroom": 4.5550524348145895,
+    "producer_batch_seconds": 0.06710722303901662,
+    "producer_p95_seconds": 0.22168986599717755,
+    "gpu_update_seconds": 0.9547112005235708,
+    "producer_batches_per_second": 14.9015255692907,
+    "gpu_batches_per_second": 1.0474371720490891,
+    "producer_headroom": 14.226653365890211,
     "sustainable": true,
-    "buffer_seconds": 60.68429971470607,
+    "buffer_seconds": 60.146805632984965,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435289.3554843971,
-    "upload_bytes_per_second": 5460147.13621773,
+    "generation_bytes_per_second": 371137.6150884591,
+    "upload_bytes_per_second": 5469356.660491981,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.4029626693624,
+    "estimated_full_shards_per_hour": 79.63749255647974,
     "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": 1.3927226420000807,
-      "bytes": 7163586,
-      "pages": 36,
-      "shards": 4
-    },
     {
       "seconds": 1.230830958000297,
       "bytes": 6020552,
@@ -562,6 +556,12 @@
       "bytes": 12535029,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.2274471419987094,
+      "bytes": 6988556,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790956870018829698-ff939fa5/train.jsonl

@@ -213,3 +213,4 @@
 {"step": 546300, "loss": 0.5578228831291199, "lr": 0.0003, "grad_norm": 0.14084938995452082, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79961161668, "tokens_per_second": 152438.24359266125, "data_wait_s": 1.0360003216192126e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 546400, "loss": 0.5712135434150696, "lr": 0.0003, "grad_norm": 0.17193361680797545, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79975666044, "tokens_per_second": 152001.5084784445, "data_wait_s": 9.99399708234705e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 546500, "loss": 0.5454045534133911, "lr": 0.0003, "grad_norm": 0.17001216940801897, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 79990366382, "tokens_per_second": 135680.14544971648, "data_wait_s": 1.2560998584376648e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 546600, "loss": 0.5690239071846008, "lr": 0.0003, "grad_norm": 0.15008448037025998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 80005064298, "tokens_per_second": 153047.96694378424, "data_wait_s": 1.503400198998861e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790956870.abdf5fd7-18c6-4e42-b51e-c994b21d5ba1-dh6p5.3284.0.1790956870018829698-ff939fa5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5023f16228847120da3b7246ccb61fa3d80e9cb6e760368ede5da68f2891227a
-size 458667
+oid sha256:6a957ca4b4b969322f26c39347e1e2ee74becc7ebf1a7eb09c5b2485400939ab
+size 460785