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": 75456.87511841908,
-      "required_pages_per_second_upper_estimate": 0.0065404106915630126
+      "required_token_positions_per_second": 75438.24172883222,
+      "required_pages_per_second_upper_estimate": 0.0065387955965795184
     },
     {
       "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": 9432.109389802385,
-      "required_pages_per_second_upper_estimate": 0.003762609458194664
+      "required_token_positions_per_second": 9429.780216104027,
+      "required_pages_per_second_upper_estimate": 0.0037616803159821396
     },
     {
       "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": 18864.21877960477,
-      "required_pages_per_second_upper_estimate": 0.004363905015750213
+      "required_token_positions_per_second": 18859.560432208054,
+      "required_pages_per_second_upper_estimate": 0.0043628273890640975
     },
     {
       "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": 28296.328169407152,
-      "required_pages_per_second_upper_estimate": 0.0017230979261565345
+      "required_token_positions_per_second": 28289.34064831208,
+      "required_pages_per_second_upper_estimate": 0.0017226724227825346
     },
     {
       "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": 28296.328169407152,
-      "required_pages_per_second_upper_estimate": 0.0021783296717306304
+      "required_token_positions_per_second": 28289.34064831208,
+      "required_pages_per_second_upper_estimate": 0.0021777917530140358
     },
     {
       "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": 28296.328169407152,
-      "required_pages_per_second_upper_estimate": 0.3513456941456368
+      "required_token_positions_per_second": 28289.34064831208,
+      "required_pages_per_second_upper_estimate": 0.3512589325193648
     }
   ],
   "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": 0.0840861957577772,
-    "producer_p95_seconds": 0.28361618499911856,
-    "gpu_update_seconds": 0.9770879046381679,
-    "producer_batches_per_second": 11.892558475122941,
-    "gpu_batches_per_second": 1.02344936955321,
-    "producer_headroom": 11.620075041244759,
+    "producer_batch_seconds": 0.09697521014820154,
+    "producer_p95_seconds": 0.28277453600094304,
+    "gpu_update_seconds": 0.9773292472141677,
+    "producer_batches_per_second": 10.311913719720312,
+    "gpu_batches_per_second": 1.023196638032121,
+    "producer_headroom": 10.0781348730317,
     "sustainable": true,
-    "buffer_seconds": 60.57945008756641,
+    "buffer_seconds": 60.594413327278396,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 373228.7193388393,
-    "upload_bytes_per_second": 6401005.334243602,
+    "generation_bytes_per_second": 373334.968358848,
+    "upload_bytes_per_second": 6333530.1283394545,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 80.08619485019574,
+    "estimated_full_shards_per_hour": 80.10899341653901,
     "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": 1.1390384229998745,
-      "bytes": 12216394,
-      "pages": 44,
-      "shards": 4
-    },
     {
       "seconds": 0.9729804899998271,
       "bytes": 6782943,
@@ -562,6 +556,12 @@
       "bytes": 7099055,
       "pages": 43,
       "shards": 2
+    },
+    {
+      "seconds": 1.0609590189997107,
+      "bytes": 7221766,
+      "pages": 30,
+      "shards": 3
     }
   ],
   "cache_key": {

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

@@ -163,3 +163,4 @@
 {"step": 56300, "loss": 0.6548167616128922, "lr": 0.0003, "grad_norm": 0.13857109300162795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8240281188, "tokens_per_second": 146028.79511833427, "data_wait_s": 8.115999662550166e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 56400, "loss": 0.6107414867728949, "lr": 0.0003, "grad_norm": 0.14392933913865694, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8254997053, "tokens_per_second": 150781.31389081103, "data_wait_s": 1.1790001735789701e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 56500, "loss": 0.621210690587759, "lr": 0.0003, "grad_norm": 0.12846616828292443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8269642041, "tokens_per_second": 149535.99882208108, "data_wait_s": 1.164700006484054e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 56600, "loss": 0.6294011455029249, "lr": 0.0003, "grad_norm": 0.13477242866264275, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 8284336702, "tokens_per_second": 150632.95491428688, "data_wait_s": 1.138600055128336e-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 sha256:9145e4dfcbcedf4da3055e3155c6242a40f25a8664b418a74aafe1fdac3f0014
-size 351984
+oid sha256:adc50a0c251b9fdb5054f4bb2e7ff357a245dde2c23cb6cd3f075a1108169248
+size 354102