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": 77275.6032103166,
-      "required_pages_per_second_upper_estimate": 0.005346105534192435
+      "required_token_positions_per_second": 77388.76318187494,
+      "required_pages_per_second_upper_estimate": 0.0053539342035921495
     },
     {
       "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": 9659.450401289576,
-      "required_pages_per_second_upper_estimate": 0.004044210170758101
+      "required_token_positions_per_second": 9673.595397734367,
+      "required_pages_per_second_upper_estimate": 0.00405013238538842
     },
     {
       "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": 19318.90080257915,
-      "required_pages_per_second_upper_estimate": 0.004023617563502452
+      "required_token_positions_per_second": 19347.190795468734,
+      "required_pages_per_second_upper_estimate": 0.00402950962296407
     },
     {
       "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": 28978.35120386873,
-      "required_pages_per_second_upper_estimate": 0.0017372079072916484
+      "required_token_positions_per_second": 29020.786193203105,
+      "required_pages_per_second_upper_estimate": 0.0017397518201077672
     },
     {
       "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": 28978.35120386873,
-      "required_pages_per_second_upper_estimate": 0.0022529393438044786
+      "required_token_positions_per_second": 29020.786193203105,
+      "required_pages_per_second_upper_estimate": 0.002256238477573434
     },
     {
       "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": 28978.35120386873,
-      "required_pages_per_second_upper_estimate": 0.3717031747138791
+      "required_token_positions_per_second": 29020.786193203105,
+      "required_pages_per_second_upper_estimate": 0.3722474851939188
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.10766386294551467,
-    "producer_p95_seconds": 0.32867089600040345,
-    "gpu_update_seconds": 0.9540915494291089,
-    "producer_batches_per_second": 9.28816756747869,
-    "gpu_batches_per_second": 1.0481174480565947,
-    "producer_headroom": 8.86176218581294,
+    "producer_batch_seconds": 0.11156282515605653,
+    "producer_p95_seconds": 0.30862537599750794,
+    "gpu_update_seconds": 0.9526964506039254,
+    "producer_batches_per_second": 8.963559309305568,
+    "gpu_batches_per_second": 1.0496522783999964,
+    "producer_headroom": 8.539551138753188,
     "sustainable": true,
-    "buffer_seconds": 60.10776761403386,
+    "buffer_seconds": 60.0198763880473,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 535253.9099393168,
+    "generation_bytes_per_second": 535823.756870315,
     "upload_bytes_per_second": 5961717.847362944,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 114.85302899965885,
+    "estimated_full_shards_per_hour": 114.97530488569342,
     "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": [

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

@@ -314,3 +314,4 @@
 {"step": 516400, "loss": 0.5965888500213623, "lr": 0.0003, "grad_norm": 0.1514518811000016, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75583191886, "tokens_per_second": 153272.14447573066, "data_wait_s": 2.549700002418831e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 516500, "loss": 0.5942535400390625, "lr": 0.0003, "grad_norm": 0.16532523981230376, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75597973591, "tokens_per_second": 153758.60466063625, "data_wait_s": 2.15519976336509e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 516600, "loss": 0.5852409601211548, "lr": 0.0003, "grad_norm": 0.16251029845161472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75612706432, "tokens_per_second": 153456.53370417174, "data_wait_s": 1.7587997717782855e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 516700, "loss": 0.5686122179031372, "lr": 0.0003, "grad_norm": 0.16759162474548717, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 75627365397, "tokens_per_second": 152800.1211876946, "data_wait_s": 1.776799763320014e-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 sha256:3e1329b59007b3af0dc068254a6efe8d4d71b49792891005ce382182579b803e
-size 674151
+oid sha256:fd5b149883b051789a25687f2eef953fb8e002f64da72e4efc289b98765e21f0
+size 676269