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": 77515.81336660638,
-      "required_pages_per_second_upper_estimate": 0.005362723830168914
+      "required_token_positions_per_second": 77366.93190095852,
+      "required_pages_per_second_upper_estimate": 0.0053524238649228
     },
     {
       "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": 9689.476670825798,
-      "required_pages_per_second_upper_estimate": 0.004056781542793108
+      "required_token_positions_per_second": 9670.866487619815,
+      "required_pages_per_second_upper_estimate": 0.00404898984770958
     },
     {
       "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": 19378.953341651595,
-      "required_pages_per_second_upper_estimate": 0.004036124923699288
+      "required_token_positions_per_second": 19341.73297523963,
+      "required_pages_per_second_upper_estimate": 0.004028372902942745
     },
     {
       "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": 29068.430012477394,
-      "required_pages_per_second_upper_estimate": 0.0017426079943253612
+      "required_token_positions_per_second": 29012.59946285944,
+      "required_pages_per_second_upper_estimate": 0.001739261038124078
     },
     {
       "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": 29068.430012477394,
-      "required_pages_per_second_upper_estimate": 0.0022599425749589967
+      "required_token_positions_per_second": 29012.59946285944,
+      "required_pages_per_second_upper_estimate": 0.0022556019952988354
     },
     {
       "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": 29068.430012477394,
-      "required_pages_per_second_upper_estimate": 0.3728586089516219
+      "required_token_positions_per_second": 29012.59946285944,
+      "required_pages_per_second_upper_estimate": 0.3721424746072965
     }
   ],
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.0747855775624231,
-    "producer_p95_seconds": 0.20644337199883012,
-    "gpu_update_seconds": 0.9511349594089641,
-    "producer_batches_per_second": 13.371562172737192,
-    "gpu_batches_per_second": 1.0513755068170354,
-    "producer_headroom": 12.718160244400828,
+    "producer_batch_seconds": 0.09743705696887162,
+    "producer_p95_seconds": 0.21227939999880618,
+    "gpu_update_seconds": 0.952965281011571,
+    "producer_batches_per_second": 10.263035759787693,
+    "gpu_batches_per_second": 1.0493561727018026,
+    "producer_headroom": 9.78031675685788,
     "sustainable": true,
-    "buffer_seconds": 60.8726374021737,
+    "buffer_seconds": 60.03681270372897,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 369808.39453087514,
+    "generation_bytes_per_second": 369980.64322516276,
     "upload_bytes_per_second": 5415588.325594185,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 79.35227276749316,
+    "estimated_full_shards_per_hour": 79.3892333275429,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -159,3 +159,4 @@
 {"step": 573000, "loss": 0.5801893472671509, "lr": 0.0003, "grad_norm": 0.1543706092311972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83870326717, "tokens_per_second": 154341.376915835, "data_wait_s": 1.0422001651022583e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 573100, "loss": 0.5934357643127441, "lr": 0.0003, "grad_norm": 0.1694824302473004, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83885041144, "tokens_per_second": 144186.33631227972, "data_wait_s": 1.0832998668774962e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 573200, "loss": 0.5817840695381165, "lr": 0.0003, "grad_norm": 0.17621079851627106, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83899574908, "tokens_per_second": 151516.6430221442, "data_wait_s": 1.6834999769343995e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 573300, "loss": 0.6154465079307556, "lr": 0.0003, "grad_norm": 0.16023500949454553, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83913998030, "tokens_per_second": 150727.05444188064, "data_wait_s": 1.0301999282091856e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fe6fc0f2e86fb179f432c475eeb77b624dee3ed35dd17cfaa58fd546d84f7115
-size 343512
+oid sha256:a92cea1319cc0d075c66c8c4a9370def538d3dbb73a556ceead2226f38fa3fbf
+size 345630