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": 75587.7969737047,
-      "required_pages_per_second_upper_estimate": 0.0065517586661607615
+      "required_token_positions_per_second": 75614.46905164768,
+      "required_pages_per_second_upper_estimate": 0.006554070534277096
     },
     {
       "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": 9448.474621713087,
-      "required_pages_per_second_upper_estimate": 0.003769137793885865
+      "required_token_positions_per_second": 9451.80863145596,
+      "required_pages_per_second_upper_estimate": 0.0037704677802201853
     },
     {
       "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": 18896.949243426174,
-      "required_pages_per_second_upper_estimate": 0.004371476632518806
+      "required_token_positions_per_second": 18903.61726291192,
+      "required_pages_per_second_upper_estimate": 0.004373019161473699
     },
     {
       "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": 28345.423865139262,
-      "required_pages_per_second_upper_estimate": 0.001726087596441412
+      "required_token_positions_per_second": 28355.425894367883,
+      "required_pages_per_second_upper_estimate": 0.0017266966675448432
     },
     {
       "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": 28345.423865139262,
-      "required_pages_per_second_upper_estimate": 0.0021821091942936725
+      "required_token_positions_per_second": 28355.425894367883,
+      "required_pages_per_second_upper_estimate": 0.002182879178191081
     },
     {
       "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": 28345.423865139262,
-      "required_pages_per_second_upper_estimate": 0.35195529837390593
+      "required_token_positions_per_second": 28355.425894367883,
+      "required_pages_per_second_upper_estimate": 0.35207949010228695
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.12198301791369204,
-    "producer_p95_seconds": 0.37699189400154864,
-    "gpu_update_seconds": 0.9753955393838019,
-    "producer_batches_per_second": 8.197862432847339,
-    "gpu_batches_per_second": 1.0252251108629653,
-    "producer_headroom": 7.996158449481337,
+    "producer_batch_seconds": 0.1586184294690156,
+    "producer_p95_seconds": 0.3814739430017653,
+    "gpu_update_seconds": 0.9750514805524965,
+    "producer_batches_per_second": 6.304437658017155,
+    "gpu_batches_per_second": 1.0255868740729124,
+    "producer_headroom": 6.14715127250054,
     "sustainable": true,
-    "buffer_seconds": 60.47452344179572,
+    "buffer_seconds": 60.45319179425478,
     "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": 286113.3896229766,
-    "upload_bytes_per_second": 5180145.650872773,
+    "generation_bytes_per_second": 286276.32369500207,
+    "upload_bytes_per_second": 5160619.877454572,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.39327303425763,
+    "estimated_full_shards_per_hour": 61.428234893203225,
     "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": 1.4158115079990239,
-      "bytes": 6180346,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.5717087800003355,
       "bytes": 560796,
@@ -562,6 +556,12 @@
       "bytes": 5831579,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.0129043119959533,
+      "bytes": 2534995,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -401,3 +401,4 @@
 {"step": 277200, "loss": 0.672737043350935, "lr": 0.0003, "grad_norm": 0.19240088435973923, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40574057787, "tokens_per_second": 123632.47096469428, "data_wait_s": 1.122899993788451e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 277300, "loss": 0.5807579811662436, "lr": 0.0003, "grad_norm": 0.15799625978100793, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40588611599, "tokens_per_second": 147551.60372950922, "data_wait_s": 1.6421996406279504e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 277400, "loss": 0.5671660769730806, "lr": 0.0003, "grad_norm": 0.14708343725716377, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40603211126, "tokens_per_second": 148392.47630144472, "data_wait_s": 1.7530997865833342e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 277500, "loss": 0.6091627143323421, "lr": 0.0003, "grad_norm": 0.15208406848093192, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40617975581, "tokens_per_second": 149867.00935391747, "data_wait_s": 1.4083998394198716e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:71dc716a1fc8d80138fa865bf696e830831a7e52bae973dbc5c9c12606e319f6
-size 859983
+oid sha256:c101d32a0c4ceca2d2f8292dbf4b045fb814532d7808b18c0bec72d340d86e86
+size 862101