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": 75749.20111174913,
-      "required_pages_per_second_upper_estimate": 0.005240505495272712
+      "required_token_positions_per_second": 76069.56968206368,
+      "required_pages_per_second_upper_estimate": 0.00526266933632457
     },
     {
       "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": 9468.650138968642,
-      "required_pages_per_second_upper_estimate": 0.00396432608528688
+      "required_token_positions_per_second": 9508.69621025796,
+      "required_pages_per_second_upper_estimate": 0.003981092539078654
     },
     {
       "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": 18937.300277937284,
-      "required_pages_per_second_upper_estimate": 0.003944140237702126
+      "required_token_positions_per_second": 19017.39242051592,
+      "required_pages_per_second_upper_estimate": 0.003960821318829419
     },
     {
       "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": 28405.950416905926,
-      "required_pages_per_second_upper_estimate": 0.0017028933541186236
+      "required_token_positions_per_second": 28526.088630773877,
+      "required_pages_per_second_upper_estimate": 0.001710095456599576
     },
     {
       "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": 28405.950416905926,
-      "required_pages_per_second_upper_estimate": 0.002208437700343101
+      "required_token_positions_per_second": 28526.088630773877,
+      "required_pages_per_second_upper_estimate": 0.002217777918039178
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28405.950416905926,
-      "required_pages_per_second_upper_estimate": 0.3643610320148013
+      "required_token_positions_per_second": 28526.088630773877,
+      "required_pages_per_second_upper_estimate": 0.3659020360279355
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.1007617098358935,
-    "producer_p95_seconds": 0.3085167899989756,
-    "gpu_update_seconds": 0.9733171956656367,
-    "producer_batches_per_second": 9.924404832238947,
-    "gpu_batches_per_second": 1.0274142945929516,
-    "producer_headroom": 9.659593879965305,
+    "producer_batch_seconds": 0.3580228435393451,
+    "producer_p95_seconds": 0.2806917589987279,
+    "gpu_update_seconds": 0.9692180501105715,
+    "producer_batches_per_second": 2.7931178639725665,
+    "gpu_batches_per_second": 1.0317595714255599,
+    "producer_headroom": 2.7071402498484956,
     "sustainable": true,
-    "buffer_seconds": 60.345666131269475,
+    "buffer_seconds": 60.09151910685543,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 333869.1203607204,
+    "generation_bytes_per_second": 309593.29874894355,
     "upload_bytes_per_second": 5407269.380927152,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 71.64054115406235,
+    "estimated_full_shards_per_hour": 66.4315149483798,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -264,3 +264,4 @@
 {"step": 746400, "loss": 0.5150867700576782, "lr": 5e-05, "grad_norm": 0.2449501627606696, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109252631258, "tokens_per_second": 150056.3392223231, "data_wait_s": 1.0732001101132482e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 746500, "loss": 0.535110592842102, "lr": 5e-05, "grad_norm": 0.2323000257424983, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109267363758, "tokens_per_second": 151393.53563973712, "data_wait_s": 1.2826003512600437e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 746600, "loss": 0.5292147397994995, "lr": 5e-05, "grad_norm": 0.23191915560626236, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109282065921, "tokens_per_second": 152192.27625468437, "data_wait_s": 1.2601001799339429e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 746700, "loss": 0.5337346196174622, "lr": 5e-05, "grad_norm": 0.2351994291797505, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109296600270, "tokens_per_second": 149347.83223287825, "data_wait_s": 1.1419000657042488e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ff5ec080ab1362bad5b1a746300e0bd428aad009805069a3b239ed556e07a47d
-size 567468
+oid sha256:3ceecadb8a71ef36adffd6aaf7a7545bf66301a9ca331e1d15569df5b84668bc
+size 569586