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": 75985.52004923989,
-      "required_pages_per_second_upper_estimate": 0.006586232294328233
+      "required_token_positions_per_second": 76119.10653187743,
+      "required_pages_per_second_upper_estimate": 0.006597811232071411
     },
     {
       "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": 9498.190006154986,
-      "required_pages_per_second_upper_estimate": 0.0037889700040509757
+      "required_token_positions_per_second": 9514.88831648468,
+      "required_pages_per_second_upper_estimate": 0.0037956312097034784
     },
     {
       "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": 18996.38001230997,
-      "required_pages_per_second_upper_estimate": 0.004394478190978307
+      "required_token_positions_per_second": 19029.77663296936,
+      "required_pages_per_second_upper_estimate": 0.004402203911407407
     },
     {
       "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": 28494.570018464958,
-      "required_pages_per_second_upper_estimate": 0.0017351698146695557
+      "required_token_positions_per_second": 28544.66494945404,
+      "required_pages_per_second_upper_estimate": 0.0017382203331390003
     },
     {
       "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": 28494.570018464958,
-      "required_pages_per_second_upper_estimate": 0.0021935908780396614
+      "required_token_positions_per_second": 28544.66494945404,
+      "required_pages_per_second_upper_estimate": 0.0021974473244988453
     },
     {
       "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": 28494.570018464958,
-      "required_pages_per_second_upper_estimate": 0.3538071944381459
+      "required_token_positions_per_second": 28544.66494945404,
+      "required_pages_per_second_upper_estimate": 0.35442920582408133
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.4526491880314154,
-    "producer_p95_seconds": 0.3263963059980597,
-    "gpu_update_seconds": 0.9702901283326484,
-    "producer_batches_per_second": 2.209216378690591,
-    "gpu_batches_per_second": 1.0306195753206366,
-    "producer_headroom": 2.1435808435942825,
+    "producer_batch_seconds": 0.10268126689058477,
+    "producer_p95_seconds": 0.3414515370022855,
+    "gpu_update_seconds": 0.9685873016536777,
+    "producer_batches_per_second": 9.738874775138695,
+    "gpu_batches_per_second": 1.0324314579518967,
+    "producer_headroom": 9.432950439594656,
     "sustainable": true,
-    "buffer_seconds": 60.1579879566242,
+    "buffer_seconds": 60.05241270252802,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 298158.2191129906,
+    "generation_bytes_per_second": 299325.53843147174,
     "upload_bytes_per_second": 3788275.8141807066,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.9778130535344,
+    "estimated_full_shards_per_hour": 64.22829260547746,
     "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-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -294,3 +294,4 @@
 {"step": 69400, "loss": 0.5570238828659058, "lr": 0.0003, "grad_norm": 0.13479370695657739, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10157305576, "tokens_per_second": 150245.46984794692, "data_wait_s": 1.567399885971099e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 69500, "loss": 0.6252729836851358, "lr": 0.0003, "grad_norm": 0.13367581634287012, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10171816427, "tokens_per_second": 148491.94474402617, "data_wait_s": 1.2938002328155562e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 69600, "loss": 0.5945924948900938, "lr": 0.0003, "grad_norm": 0.14415718360304616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10186342641, "tokens_per_second": 148896.37974293946, "data_wait_s": 1.3379001757130027e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 69700, "loss": 0.6128690466284752, "lr": 0.0003, "grad_norm": 0.1299160860011532, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 10201051763, "tokens_per_second": 150101.95188452434, "data_wait_s": 1.0831001418409869e-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:e70dae05d5c99e15a62e03027cc48abaee8dc5c9d01b23e2cfa8e8806dfa704b
-size 631008
+oid sha256:1546671dc6ccdc274789e79eb85f0e082e00cbe937d0f3bdd2574046cc74e80c
+size 633126