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": 76902.07767355378,
-      "required_pages_per_second_upper_estimate": 0.00532026417086053
+      "required_token_positions_per_second": 76857.85348149325,
+      "required_pages_per_second_upper_estimate": 0.00531720463863952
     },
     {
       "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": 9612.759709194223,
-      "required_pages_per_second_upper_estimate": 0.004024661752990299
+      "required_token_positions_per_second": 9607.231685186656,
+      "required_pages_per_second_upper_estimate": 0.0040223472847766
     },
     {
       "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": 19225.519418388445,
-      "required_pages_per_second_upper_estimate": 0.004004168683808235
+      "required_token_positions_per_second": 19214.463370373313,
+      "required_pages_per_second_upper_estimate": 0.004001866000574276
     },
     {
       "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": 28838.279127582668,
-      "required_pages_per_second_upper_estimate": 0.0017288108001959756
+      "required_token_positions_per_second": 28821.69505555997,
+      "required_pages_per_second_upper_estimate": 0.0017278166104006266
     },
     {
       "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": 28838.279127582668,
-      "required_pages_per_second_upper_estimate": 0.0022420493559851882
+      "required_token_positions_per_second": 28821.69505555997,
+      "required_pages_per_second_upper_estimate": 0.0022407600173310472
     },
     {
       "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": 28838.279127582668,
-      "required_pages_per_second_upper_estimate": 0.36990648051695935
+      "required_token_positions_per_second": 28821.69505555997,
+      "required_pages_per_second_upper_estimate": 0.36969375784764136
     }
   ],
   "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": 0.09816033015621883,
+    "producer_batch_seconds": 0.28076332903924595,
     "producer_p95_seconds": 0.28479360900018946,
-    "gpu_update_seconds": 0.9587257227687969,
-    "producer_batches_per_second": 10.187414797897826,
-    "gpu_batches_per_second": 1.0430511837233314,
-    "producer_headroom": 9.76693661526013,
+    "gpu_update_seconds": 0.9592773758345087,
+    "producer_batches_per_second": 3.5617187024457064,
+    "gpu_batches_per_second": 1.042451354729455,
+    "producer_headroom": 3.4166761703428086,
     "sustainable": true,
-    "buffer_seconds": 60.3997205344342,
+    "buffer_seconds": 60.434474677574045,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 429130.40289111226,
+    "generation_bytes_per_second": 444683.24869538116,
     "upload_bytes_per_second": 5048309.317896662,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.08139481592202,
+    "estimated_full_shards_per_hour": 95.41867347379757,
     "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-v1/1791039222312944374-63c10d34/train.jsonl

@@ -262,3 +262,4 @@
 {"step": 630600, "loss": 0.5384798645973206, "lr": 0.0001, "grad_norm": 0.16188742320042304, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92302896849, "tokens_per_second": 151042.50398817108, "data_wait_s": 1.0980998922605067e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 630700, "loss": 0.5708792209625244, "lr": 0.0001, "grad_norm": 0.17876038448995485, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92317570685, "tokens_per_second": 151798.81976997806, "data_wait_s": 1.4925000868970528e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 630800, "loss": 0.5411971807479858, "lr": 0.0001, "grad_norm": 0.16672516889319627, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92332189115, "tokens_per_second": 151532.8481168999, "data_wait_s": 1.1815998732345179e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 630900, "loss": 0.5639351010322571, "lr": 0.0001, "grad_norm": 0.1630223324654458, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 92346870334, "tokens_per_second": 152068.6482738499, "data_wait_s": 1.6493999282829463e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1791039222.f2a393d4-e1e4-407b-a6aa-034aa2ada5ac-vwfgg.353.0.1791039222312944374-63c10d34

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4cb2b87b138c9d80daa54c9deced9d88281c94834fcf6dd072033fca7c6d753f
-size 564015
+oid sha256:58816f9a20164128f00a12bc49a54b9b4ebe3aaa83fdee238fed0bc5c21b54d6
+size 566133