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": 76261.13710414592,
-      "required_pages_per_second_upper_estimate": 0.005275922415081878
+      "required_token_positions_per_second": 76694.3385448136,
+      "required_pages_per_second_upper_estimate": 0.005305892295913096
     },
     {
       "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": 9532.64213801824,
-      "required_pages_per_second_upper_estimate": 0.00399111819898405
+      "required_token_positions_per_second": 9586.7923181017,
+      "required_pages_per_second_upper_estimate": 0.004013789748600649
     },
     {
       "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": 19065.28427603648,
-      "required_pages_per_second_upper_estimate": 0.003970795929341189
+      "required_token_positions_per_second": 19173.5846362034,
+      "required_pages_per_second_upper_estimate": 0.003993352038291398
     },
     {
       "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": 28597.92641405472,
-      "required_pages_per_second_upper_estimate": 0.0017144020220173188
+      "required_token_positions_per_second": 28760.376954305095,
+      "required_pages_per_second_upper_estimate": 0.0017241406838577162
     },
     {
       "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": 28597.92641405472,
-      "required_pages_per_second_upper_estimate": 0.002223362989708249
+      "required_token_positions_per_second": 28760.376954305095,
+      "required_pages_per_second_upper_estimate": 0.0022359928046683042
     },
     {
       "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": 28597.92641405472,
-      "required_pages_per_second_upper_estimate": 0.36682349397846
+      "required_token_positions_per_second": 28760.376954305095,
+      "required_pages_per_second_upper_estimate": 0.36890723508299145
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3491371905000129,
-    "producer_p95_seconds": 0.15919983200001298,
-    "gpu_update_seconds": 0.9667833814137003,
-    "producer_batches_per_second": 2.8642036059460216,
-    "gpu_batches_per_second": 1.0343578708787153,
-    "producer_headroom": 2.7690644472138084,
+    "producer_batch_seconds": 0.049116768054719984,
+    "producer_p95_seconds": 0.1561628380000002,
+    "gpu_update_seconds": 0.9613225878063957,
+    "producer_batches_per_second": 20.35964579114653,
+    "gpu_batches_per_second": 1.0402335414606878,
+    "producer_headroom": 19.572187378766575,
     "sustainable": true,
-    "buffer_seconds": 60.907353029063124,
+    "buffer_seconds": 60.56332303180293,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 323911.1311481798,
-    "upload_bytes_per_second": 431426.06842612405,
+    "generation_bytes_per_second": 323291.5210563767,
+    "upload_bytes_per_second": 398433.02046541986,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 69.50378847917601,
+    "estimated_full_shards_per_hour": 69.37083457725979,
     "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": [
@@ -190,6 +190,12 @@
       "bytes": 1106661,
       "pages": 32,
       "shards": 1
+    },
+    {
+      "seconds": 33.80818266799997,
+      "bytes": 13381894,
+      "pages": 39,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -34,3 +34,4 @@
 {"step": 648400, "loss": 0.6239676475524902, "lr": 0.0001, "grad_norm": 0.21783978781609148, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94909152120, "tokens_per_second": 152245.14727116865, "data_wait_s": 1.3509999917005189e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 648500, "loss": 0.5002675652503967, "lr": 0.0001, "grad_norm": 0.17678204568389247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94923748418, "tokens_per_second": 150892.88864365392, "data_wait_s": 1.4455999917117879e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 648600, "loss": 0.5634985566139221, "lr": 0.0001, "grad_norm": 0.17079177724091404, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94938416805, "tokens_per_second": 152800.09515960494, "data_wait_s": 1.3675999980478082e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 648700, "loss": 0.546423614025116, "lr": 0.0001, "grad_norm": 0.1903061926083244, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 94953087129, "tokens_per_second": 152879.50500266466, "data_wait_s": 1.6120000054797856e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1254c214129ef1eb00992149b19685eea18e6afd4e1fd97ffb94874bcf0bc1ac
-size 76413
+oid sha256:f55492b08ac8d0d248d3976f7b145897ec3deb8669bde3f76045e11a93ef1052
+size 78531