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": 77410.3561746727,
-      "required_pages_per_second_upper_estimate": 0.005355428056936545
+      "required_token_positions_per_second": 77211.82279985609,
+      "required_pages_per_second_upper_estimate": 0.005341693057405835
     },
     {
       "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": 9676.294521834088,
-      "required_pages_per_second_upper_estimate": 0.004051262452284853
+      "required_token_positions_per_second": 9651.477849982011,
+      "required_pages_per_second_upper_estimate": 0.004040872230011427
     },
     {
       "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": 19352.589043668177,
-      "required_pages_per_second_upper_estimate": 0.004030633935702636
+      "required_token_positions_per_second": 19302.955699964023,
+      "required_pages_per_second_upper_estimate": 0.004020296619128355
     },
     {
       "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": 29028.883565502267,
-      "required_pages_per_second_upper_estimate": 0.0017402372452131303
+      "required_token_positions_per_second": 28954.433549946036,
+      "required_pages_per_second_upper_estimate": 0.001735774080459126
     },
     {
       "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": 29028.883565502267,
-      "required_pages_per_second_upper_estimate": 0.0022568680126531173
+      "required_token_positions_per_second": 28954.433549946036,
+      "required_pages_per_second_upper_estimate": 0.0022510798514146307
     },
     {
       "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": 29028.883565502267,
-      "required_pages_per_second_upper_estimate": 0.37235134959149147
+      "required_token_positions_per_second": 28954.433549946036,
+      "required_pages_per_second_upper_estimate": 0.37139638473013475
     }
   ],
   "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": 0.09675821350788283,
-    "producer_p95_seconds": 0.26541237399942474,
-    "gpu_update_seconds": 0.9524307036339731,
-    "producer_batches_per_second": 10.33503992835224,
-    "gpu_batches_per_second": 1.0499451521087335,
-    "producer_headroom": 9.84340935104573,
+    "producer_batch_seconds": 0.11323504734383505,
+    "producer_p95_seconds": 0.2793388550016971,
+    "gpu_update_seconds": 0.9548796716160082,
+    "producer_batches_per_second": 8.831188076987578,
+    "gpu_batches_per_second": 1.0472523708747843,
+    "producer_headroom": 8.432721970933105,
     "sustainable": true,
-    "buffer_seconds": 60.003134328940305,
+    "buffer_seconds": 60.15741931180852,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 308683.0631269223,
+    "generation_bytes_per_second": 308710.87964071456,
     "upload_bytes_per_second": 4703183.91493153,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.23619957309486,
+    "estimated_full_shards_per_hour": 66.24216834941937,
     "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-v1/1791143623306220418-eb05a537/train.jsonl

@@ -291,3 +291,4 @@
 {"step": 714100, "loss": 0.5695791840553284, "lr": 8e-05, "grad_norm": 0.23223308443294916, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104524641161, "tokens_per_second": 154463.46389232014, "data_wait_s": 9.605999366613105e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 714200, "loss": 0.5603662133216858, "lr": 8e-05, "grad_norm": 0.20530037538483623, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104539331716, "tokens_per_second": 152757.4018076974, "data_wait_s": 1.34839974634815e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 714300, "loss": 0.5359557867050171, "lr": 8e-05, "grad_norm": 0.208079726218237, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104553813379, "tokens_per_second": 150972.46325048085, "data_wait_s": 1.2636002793442458e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 714400, "loss": 0.51793372631073, "lr": 8e-05, "grad_norm": 0.22320855063949332, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104568324228, "tokens_per_second": 150712.13782569245, "data_wait_s": 1.3154000043869019e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a9393295f4ac30812107999049d60ca2b6f3fbc5ad9bd17aaaa9284d69192a1
-size 624654
+oid sha256:9c179571f86bf5c0ab5656ea3ca6f1a5e5d0deda4ce79593f4c474f1620def4e
+size 626772