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": 77706.6146800692,
-      "required_pages_per_second_upper_estimate": 0.005375923907754314
+      "required_token_positions_per_second": 77504.13216348628,
+      "required_pages_per_second_upper_estimate": 0.0053619156974329405
     },
     {
       "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": 9713.32683500865,
-      "required_pages_per_second_upper_estimate": 0.004066767108488405
+      "required_token_positions_per_second": 9688.016520435785,
+      "required_pages_per_second_upper_estimate": 0.004056170208316217
     },
     {
       "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": 19426.6536700173,
-      "required_pages_per_second_upper_estimate": 0.004046059644155613
+      "required_token_positions_per_second": 19376.03304087157,
+      "required_pages_per_second_upper_estimate": 0.00403551670206032
     },
     {
       "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": 29139.980505025946,
-      "required_pages_per_second_upper_estimate": 0.0017468973371023704
+      "required_token_positions_per_second": 29064.049561307358,
+      "required_pages_per_second_upper_estimate": 0.001742345392966279
     },
     {
       "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": 29139.980505025946,
-      "required_pages_per_second_upper_estimate": 0.0022655053110372902
+      "required_token_positions_per_second": 29064.049561307358,
+      "required_pages_per_second_upper_estimate": 0.0022596020141480945
     },
     {
       "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": 29139.980505025946,
-      "required_pages_per_second_upper_estimate": 0.3737763818451013
+      "required_token_positions_per_second": 29064.049561307358,
+      "required_pages_per_second_upper_estimate": 0.37280242122737467
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07119491258580979,
-    "producer_p95_seconds": 0.186670270999457,
-    "gpu_update_seconds": 0.9487995366102384,
-    "producer_batches_per_second": 14.045947437532424,
-    "gpu_batches_per_second": 1.053963415257015,
-    "producer_headroom": 13.326788419982531,
+    "producer_batch_seconds": 0.0744082698827242,
+    "producer_p95_seconds": 0.17372013199928915,
+    "gpu_update_seconds": 0.9512783117741264,
+    "producer_batches_per_second": 13.43936637118579,
+    "gpu_batches_per_second": 1.0512170703597858,
+    "producer_headroom": 12.784577752895585,
     "sustainable": true,
-    "buffer_seconds": 60.72317034305526,
+    "buffer_seconds": 60.88181195354409,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 76,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 250326.16363505236,
-    "upload_bytes_per_second": 5820966.888306321,
+    "generation_bytes_per_second": 243368.26589684637,
+    "upload_bytes_per_second": 5887064.024184073,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 53.71416742123297,
+    "estimated_full_shards_per_hour": 52.22116453818362,
     "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": [
@@ -496,6 +496,12 @@
       "bytes": 5823273,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.1922360469998239,
+      "bytes": 10997251,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -67,3 +67,4 @@
 {"step": 491700, "loss": 0.5966281294822693, "lr": 0.0003, "grad_norm": 0.16710601263201885, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71968244192, "tokens_per_second": 78997.4518323443, "data_wait_s": 1.614299981156364e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 491800, "loss": 0.5912267565727234, "lr": 0.0003, "grad_norm": 0.17450495071352756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71982856035, "tokens_per_second": 153186.3163065317, "data_wait_s": 1.0395999197498895e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 491900, "loss": 0.6098073124885559, "lr": 0.0003, "grad_norm": 0.1502846487220614, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 71997510019, "tokens_per_second": 144049.5154796702, "data_wait_s": 1.22540004667826e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 492000, "loss": 0.5890203714370728, "lr": 0.0003, "grad_norm": 0.1959452605804415, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72012187687, "tokens_per_second": 153246.40589248177, "data_wait_s": 9.343999408883974e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ed520c82e214c81fef7a1c2a52f197f99721de8819c7929bf2e93b444444cbbf
-size 147090
+oid sha256:96fb91d2d7ca25db547f563ae95540d9a1a197cccf26da06c4db4c17da54489c
+size 149208