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": 76875.18403113878,
-      "required_pages_per_second_upper_estimate": 0.005318403606276412
+      "required_token_positions_per_second": 74813.26647189389,
+      "required_pages_per_second_upper_estimate": 0.005175755365219964
     },
     {
       "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": 9609.398003892347,
-      "required_pages_per_second_upper_estimate": 0.0040232542771807935
+      "required_token_positions_per_second": 9351.658308986736,
+      "required_pages_per_second_upper_estimate": 0.003915344049140676
     },
     {
       "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": 19218.796007784695,
-      "required_pages_per_second_upper_estimate": 0.004002768374687734
+      "required_token_positions_per_second": 18703.316617973473,
+      "required_pages_per_second_upper_estimate": 0.00389540761189573
     },
     {
       "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": 28828.194011677042,
-      "required_pages_per_second_upper_estimate": 0.0017282062128965137
+      "required_token_positions_per_second": 28054.97492696021,
+      "required_pages_per_second_upper_estimate": 0.001681852909404921
     },
     {
       "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": 28828.194011677042,
-      "required_pages_per_second_upper_estimate": 0.0022412652825832624
+      "required_token_positions_per_second": 28054.97492696021,
+      "required_pages_per_second_upper_estimate": 0.0021811508997778505
     },
     {
       "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": 28828.194011677042,
-      "required_pages_per_second_upper_estimate": 0.36977711947867575
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28054.97492696021,
+      "required_pages_per_second_upper_estimate": 0.35985909527789806
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.15066155442190166,
-    "producer_p95_seconds": 0.37203351300559007,
-    "gpu_update_seconds": 0.9590611187367826,
-    "producer_batches_per_second": 6.637393353845751,
-    "gpu_batches_per_second": 1.0426864153529023,
-    "producer_headroom": 6.3656658954353915,
+    "producer_batch_seconds": 0.37333076137502985,
+    "producer_p95_seconds": 0.3436705179992714,
+    "gpu_update_seconds": 0.9854936627810308,
+    "producer_batches_per_second": 2.678589881843272,
+    "gpu_batches_per_second": 1.0147198685966512,
+    "producer_headroom": 2.6397333537459344,
     "sustainable": true,
-    "buffer_seconds": 60.4208504804173,
+    "buffer_seconds": 60.115113429642875,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 306324.97276066034,
+    "generation_bytes_per_second": 287424.90948749735,
     "upload_bytes_per_second": 4971860.778744054,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.73020827402932,
+    "estimated_full_shards_per_hour": 61.674694666563894,
     "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-v1/1790824801708454555-41033672/train.jsonl

@@ -413,3 +413,4 @@
 {"step": 438400, "loss": 0.5706033110618591, "lr": 0.0003, "grad_norm": 0.15129333047272026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64169315097, "tokens_per_second": 130844.73358345087, "data_wait_s": 7.945003744680434e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 438500, "loss": 0.611498236656189, "lr": 0.0003, "grad_norm": 0.14693253937178602, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64183893067, "tokens_per_second": 151591.0330331417, "data_wait_s": 8.968003385234624e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 438600, "loss": 0.6123660802841187, "lr": 0.0003, "grad_norm": 0.15547021280987575, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64198395879, "tokens_per_second": 150770.17026214392, "data_wait_s": 1.3719996786676347e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 438700, "loss": 0.5546029210090637, "lr": 0.0003, "grad_norm": 0.15292847105504506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 64212953981, "tokens_per_second": 151153.98076852626, "data_wait_s": 9.530005627311766e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:aeddbe686ed669020565711f430baef1f88173edb3fb0c5d0295284a53c13713
-size 885399
+oid sha256:e1aaaf965ce7177990b1509473dd18a0e56383f36ad0990ba64cc8b0a864691a
+size 887517