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": 76726.77479246422,
-      "required_pages_per_second_upper_estimate": 0.00530813631078803
+      "required_token_positions_per_second": 74846.34535141522,
+      "required_pages_per_second_upper_estimate": 0.0051780438388588255
     },
     {
       "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": 9590.846849058027,
-      "required_pages_per_second_upper_estimate": 0.004015487296043828
+      "required_token_positions_per_second": 9355.793168926903,
+      "required_pages_per_second_upper_estimate": 0.00391707522865193
     },
     {
       "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": 19181.693698116054,
-      "required_pages_per_second_upper_estimate": 0.00399504094203746
+      "required_token_positions_per_second": 18711.586337853805,
+      "required_pages_per_second_upper_estimate": 0.003897129976459624
     },
     {
       "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": 28772.54054717408,
-      "required_pages_per_second_upper_estimate": 0.0017248698726774812
+      "required_token_positions_per_second": 28067.379506780704,
+      "required_pages_per_second_upper_estimate": 0.0016825965450244628
     },
     {
       "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": 28772.54054717408,
-      "required_pages_per_second_upper_estimate": 0.0022369384705118772
+      "required_token_positions_per_second": 28067.379506780704,
+      "required_pages_per_second_upper_estimate": 0.0021821153012969137
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28772.54054717408,
-      "required_pages_per_second_upper_estimate": 0.3690632565920663
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28067.379506780704,
+      "required_pages_per_second_upper_estimate": 0.36001820790883526
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.18783272229671866,
-    "producer_p95_seconds": 0.2882956439971167,
-    "gpu_update_seconds": 0.9609161886372064,
-    "producer_batches_per_second": 5.323885996926051,
-    "gpu_batches_per_second": 1.0406734862259144,
-    "producer_headroom": 5.115808240905174,
+    "producer_batch_seconds": 0.10544663335164728,
+    "producer_p95_seconds": 0.27246575100070913,
+    "gpu_update_seconds": 0.9850581167835997,
+    "producer_batches_per_second": 9.483470151817588,
+    "gpu_batches_per_second": 1.0151685296144641,
+    "producer_headroom": 9.341769248322912,
     "sustainable": true,
-    "buffer_seconds": 60.537719884144,
+    "buffer_seconds": 60.08854512379958,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
     "upload_pages_per_commit": 165,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 672379.5398290964,
+    "generation_bytes_per_second": 666047.1250430234,
     "upload_bytes_per_second": 5380735.178534467,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.13849852635703,
+    "estimated_full_shards_per_hour": 71.45910412534727,
     "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-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -247,3 +247,4 @@
 {"step": 669700, "loss": 0.5682086944580078, "lr": 0.0001, "grad_norm": 0.1946624194911147, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98027024485, "tokens_per_second": 150728.8789047591, "data_wait_s": 9.079998562810943e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 669800, "loss": 0.595058262348175, "lr": 0.0001, "grad_norm": 0.19365635579819224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98041666937, "tokens_per_second": 151575.4650209032, "data_wait_s": 1.4192002709023654e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 669900, "loss": 0.5402909517288208, "lr": 0.0001, "grad_norm": 0.1947490227177393, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98056439999, "tokens_per_second": 153219.01603945586, "data_wait_s": 1.0189996828557923e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 670000, "loss": 0.5217463970184326, "lr": 0.0001, "grad_norm": 0.18787271377557538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 98071134777, "tokens_per_second": 153469.18739889583, "data_wait_s": 9.882998710963875e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a3d7662d91ef1c81c699e364b10c1d3b642e8bdc7716f042d381b3c0b158300
-size 530679
+oid sha256:7b32f811559ef40f7cff3a238b2f6a94ccef22953feff58b0ca48c0b37ed6cca
+size 532797