Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75031.44071721716,
-      "required_pages_per_second_upper_estimate": 0.006503535115920463
+      "required_token_positions_per_second": 75957.2605760749,
+      "required_pages_per_second_upper_estimate": 0.0065837828348172716
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9378.930089652145,
-      "required_pages_per_second_upper_estimate": 0.003741395440263342
+      "required_token_positions_per_second": 9494.657572009362,
+      "required_pages_per_second_upper_estimate": 0.0037875608632556893
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18757.86017930429,
-      "required_pages_per_second_upper_estimate": 0.004339300825418131
+      "required_token_positions_per_second": 18989.315144018725,
+      "required_pages_per_second_upper_estimate": 0.004392843858036562
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28136.790268956436,
-      "required_pages_per_second_upper_estimate": 0.0017133829050497605
+      "required_token_positions_per_second": 28483.97271602809,
+      "required_pages_per_second_upper_estimate": 0.0017345244945508992
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28136.790268956436,
-      "required_pages_per_second_upper_estimate": 0.0021660480025247606
+      "required_token_positions_per_second": 28483.97271602809,
+      "required_pages_per_second_upper_estimate": 0.0021927750683628606
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28136.790268956436,
-      "required_pages_per_second_upper_estimate": 0.34936476736104444
+      "required_token_positions_per_second": 28483.97271602809,
+      "required_pages_per_second_upper_estimate": 0.35367561140876974
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10323113060164246,
-    "producer_p95_seconds": 0.3304543599988392,
-    "gpu_update_seconds": 0.9826280729150644,
-    "producer_batches_per_second": 9.687000366768137,
-    "gpu_batches_per_second": 1.0176790461862137,
-    "producer_headroom": 9.518718502724896,
+    "producer_batch_seconds": 0.40529017164851666,
+    "producer_p95_seconds": 0.32558403499933775,
+    "gpu_update_seconds": 0.9706511193377994,
+    "producer_batches_per_second": 2.4673679994076902,
+    "gpu_batches_per_second": 1.0302362816850437,
+    "producer_headroom": 2.3949535104433415,
     "sustainable": true,
-    "buffer_seconds": 60.92294052073399,
+    "buffer_seconds": 60.18036939894356,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 342558.0011618461,
-    "upload_bytes_per_second": 3434675.2051758748,
+    "generation_bytes_per_second": 333447.48346289224,
+    "upload_bytes_per_second": 3417077.663937365,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.50497270719087,
+    "estimated_full_shards_per_hour": 71.55006769099307,
     "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": 0.9578338470018934,
-      "bytes": 2306768,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 0.8912073430001328,
       "bytes": 6007286,
@@ -562,6 +556,12 @@
       "bytes": 7356201,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 1.2021151590015506,
+      "bytes": 960345,
+      "pages": 28,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -281,3 +281,4 @@
 {"step": 68100, "loss": 0.6183673404157162, "lr": 0.0003, "grad_norm": 0.14735262095656268, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9966627267, "tokens_per_second": 149102.65479761144, "data_wait_s": 1.5125999198062345e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 68200, "loss": 0.694906510412693, "lr": 0.0003, "grad_norm": 0.1386208189926631, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9981284720, "tokens_per_second": 149909.89668038773, "data_wait_s": 8.10399797046557e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 68300, "loss": 0.6336524225771427, "lr": 0.0003, "grad_norm": 0.13115137400273208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 9996018618, "tokens_per_second": 142216.557457655, "data_wait_s": 8.046001312322915e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 68400, "loss": 0.6469593495130539, "lr": 0.0003, "grad_norm": 0.12514281301613506, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10010732696, "tokens_per_second": 150178.65490609565, "data_wait_s": 1.562300167279318e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f76aa6f8affe553176e18396bf05804caf6225d6c71d6b84333020a26d9e6ea8
-size 603474
+oid sha256:3113af0073011f778688e9538eb0721abdcec7277e79c0d5e7d1661ecd2120f9
+size 605592