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": 77590.7020531781,
-      "required_pages_per_second_upper_estimate": 0.005367904803271639
+      "required_token_positions_per_second": 77505.39002491633,
+      "required_pages_per_second_upper_estimate": 0.005362002719205312
     },
     {
       "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": 9698.837756647263,
-      "required_pages_per_second_upper_estimate": 0.0040607008339448546
+      "required_token_positions_per_second": 9688.17375311454,
+      "required_pages_per_second_upper_estimate": 0.004056236038355421
     },
     {
       "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": 19397.675513294525,
-      "required_pages_per_second_upper_estimate": 0.004040024258315642
+      "required_token_positions_per_second": 19376.34750622908,
+      "required_pages_per_second_upper_estimate": 0.00403558219690128
     },
     {
       "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": 29096.513269941788,
-      "required_pages_per_second_upper_estimate": 0.0017442915427297002
+      "required_token_positions_per_second": 29064.521259343623,
+      "required_pages_per_second_upper_estimate": 0.001742373670543312
     },
     {
       "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": 29096.513269941788,
-      "required_pages_per_second_upper_estimate": 0.002262125924701658
+      "required_token_positions_per_second": 29064.521259343623,
+      "required_pages_per_second_upper_estimate": 0.0022596386865956337
     },
     {
       "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": 29096.513269941788,
-      "required_pages_per_second_upper_estimate": 0.3732188308249226
+      "required_token_positions_per_second": 29064.521259343623,
+      "required_pages_per_second_upper_estimate": 0.372808471663314
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.06030031553135018,
-    "producer_p95_seconds": 0.2028617980013223,
-    "gpu_update_seconds": 0.9502169467350516,
-    "producer_batches_per_second": 16.58366114983427,
-    "gpu_batches_per_second": 1.0523912496362047,
-    "producer_headroom": 15.758075863484216,
+    "producer_batch_seconds": 0.11027824343747739,
+    "producer_p95_seconds": 0.21076167700084625,
+    "gpu_update_seconds": 0.9512628731536997,
+    "producer_batches_per_second": 9.067971785086995,
+    "gpu_batches_per_second": 1.0512341311973243,
+    "producer_headroom": 8.626024893958538,
     "sustainable": true,
-    "buffer_seconds": 60.813884591043305,
+    "buffer_seconds": 60.88082388183678,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 440619.55935219006,
-    "upload_bytes_per_second": 5841925.60512766,
+    "generation_bytes_per_second": 439207.8708303551,
+    "upload_bytes_per_second": 5771810.300291891,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 94.54670033859516,
+    "estimated_full_shards_per_hour": 94.24378484423627,
     "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": 1.0899097739993522,
-      "bytes": 6114737,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.358950516000732,
       "bytes": 2583429,
@@ -562,6 +556,12 @@
       "bytes": 15653746,
       "pages": 42,
       "shards": 4
+    },
+    {
+      "seconds": 1.1028008130015223,
+      "bytes": 1043427,
+      "pages": 31,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -104,3 +104,4 @@
 {"step": 495400, "loss": 0.5644538998603821, "lr": 0.0003, "grad_norm": 0.16119446364049433, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72510205476, "tokens_per_second": 152302.51850732145, "data_wait_s": 1.0859999747481197e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 495500, "loss": 0.5806675553321838, "lr": 0.0003, "grad_norm": 0.17175863049179055, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72524850839, "tokens_per_second": 153326.10488982286, "data_wait_s": 1.7907999790622853e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 495600, "loss": 0.5824835300445557, "lr": 0.0003, "grad_norm": 0.1449822851167086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72539283354, "tokens_per_second": 150639.1675202296, "data_wait_s": 1.7366000975016505e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 495700, "loss": 0.5757331848144531, "lr": 0.0003, "grad_norm": 0.16169313458835852, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 72553855424, "tokens_per_second": 152365.2462637077, "data_wait_s": 1.3701001080335118e-05, "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 sha256:589fbaa4f6c29f7af7eb1854870787947a5c6255dd52c74251e33dab1388e512
-size 226239
+oid sha256:576ce3aa0328046aeed834056b20390b53e85659836df2432ce609f506d2cf41
+size 228357