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": 76621.88192517994,
-      "required_pages_per_second_upper_estimate": 0.005300879578844319
+      "required_token_positions_per_second": 76913.38374079444,
+      "required_pages_per_second_upper_estimate": 0.0053210463508259285
     },
     {
       "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": 9577.735240647493,
-      "required_pages_per_second_upper_estimate": 0.004009997739403856
+      "required_token_positions_per_second": 9614.172967599305,
+      "required_pages_per_second_upper_estimate": 0.004025253454772316
     },
     {
       "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": 19155.470481294986,
-      "required_pages_per_second_upper_estimate": 0.0039895793375263645
+      "required_token_positions_per_second": 19228.34593519861,
+      "required_pages_per_second_upper_estimate": 0.004004757372719531
     },
     {
       "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": 28733.205721942482,
-      "required_pages_per_second_upper_estimate": 0.0017225118099656472
+      "required_token_positions_per_second": 28842.518902797914,
+      "required_pages_per_second_upper_estimate": 0.0017290649682463676
     },
     {
       "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": 28733.205721942482,
-      "required_pages_per_second_upper_estimate": 0.0022338803608657315
+      "required_token_positions_per_second": 28842.518902797914,
+      "required_pages_per_second_upper_estimate": 0.0022423789798593735
     },
     {
       "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": 28733.205721942482,
-      "required_pages_per_second_upper_estimate": 0.3685587116884401
+      "required_token_positions_per_second": 28842.518902797914,
+      "required_pages_per_second_upper_estimate": 0.3699608638011046
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.13740474343754272,
-    "producer_p95_seconds": 0.31230779299949063,
-    "gpu_update_seconds": 0.9622316516839696,
-    "producer_batches_per_second": 7.2777691292335165,
-    "gpu_batches_per_second": 1.0392507856605353,
-    "producer_headroom": 7.002899809796972,
+    "producer_batch_seconds": 0.08820416791405705,
+    "producer_p95_seconds": 0.3188373259981745,
+    "gpu_update_seconds": 0.9585847925826604,
+    "producer_batches_per_second": 11.337332732103587,
+    "gpu_batches_per_second": 1.0432045320745773,
+    "producer_headroom": 10.867794745444124,
     "sustainable": true,
-    "buffer_seconds": 60.620594056090084,
+    "buffer_seconds": 60.3908419327076,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 307664.8505536682,
-    "upload_bytes_per_second": 5561794.481321636,
+    "generation_bytes_per_second": 307231.18801073957,
+    "upload_bytes_per_second": 5426797.245959924,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.01771485765013,
+    "estimated_full_shards_per_hour": 65.92466097108498,
     "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": 1.2140417409973452,
-      "bytes": 13269645,
-      "pages": 40,
-      "shards": 4
-    },
     {
       "seconds": 1.3757806130015524,
       "bytes": 5911161,
@@ -562,6 +556,12 @@
       "bytes": 7700249,
       "pages": 41,
       "shards": 3
+    },
+    {
+      "seconds": 0.8653922449957463,
+      "bytes": 888116,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -328,3 +328,4 @@
 {"step": 473000, "loss": 0.6076064705848694, "lr": 0.0003, "grad_norm": 0.16217944847790944, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69232935689, "tokens_per_second": 153637.8361666627, "data_wait_s": 1.4773999282624573e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 473100, "loss": 0.5601118206977844, "lr": 0.0003, "grad_norm": 0.1395458442601994, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69247507106, "tokens_per_second": 151075.84462362117, "data_wait_s": 1.485900065745227e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 473200, "loss": 0.5413728952407837, "lr": 0.0003, "grad_norm": 0.17737936939567697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69261949976, "tokens_per_second": 150620.79830957143, "data_wait_s": 1.229200279340148e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 473300, "loss": 0.5723682641983032, "lr": 0.0003, "grad_norm": 0.15079825006775918, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69276471092, "tokens_per_second": 151624.62627994956, "data_wait_s": 1.9396000425331295e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:df6642f7c465796ccb052d6f625d0dee1ac24fae0dc0a5be484f5e8e9c53115b
-size 703803
+oid sha256:264c8cc27c7057e78911a77cc486c8990366df8a79dd1e5c56678a45ad5d6b1d
+size 705921