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": 76810.32067657184,
-      "required_pages_per_second_upper_estimate": 0.005313916208903753
+      "required_token_positions_per_second": 76670.2398315143,
+      "required_pages_per_second_upper_estimate": 0.005304225091010331
     },
     {
       "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": 9601.29008457148,
-      "required_pages_per_second_upper_estimate": 0.004019859660673755
+      "required_token_positions_per_second": 9583.779978939287,
+      "required_pages_per_second_upper_estimate": 0.0040125285450981416
     },
     {
       "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": 19202.58016914296,
-      "required_pages_per_second_upper_estimate": 0.003999391043139083
+      "required_token_positions_per_second": 19167.559957878573,
+      "required_pages_per_second_upper_estimate": 0.003992097256677788
     },
     {
       "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": 28803.87025371444,
-      "required_pages_per_second_upper_estimate": 0.0017267480407468837
+      "required_token_positions_per_second": 28751.339936817858,
+      "required_pages_per_second_upper_estimate": 0.0017235989284580832
     },
     {
       "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": 28803.87025371444,
-      "required_pages_per_second_upper_estimate": 0.002239374217390575
+      "required_token_positions_per_second": 28751.339936817858,
+      "required_pages_per_second_upper_estimate": 0.0022352902163083107
     },
     {
       "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": 28803.87025371444,
-      "required_pages_per_second_upper_estimate": 0.36946512042834806
+      "required_token_positions_per_second": 28751.339936817858,
+      "required_pages_per_second_upper_estimate": 0.36879131792585856
     }
   ],
   "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": 0.10000629046865583,
-    "producer_p95_seconds": 0.2380408479984908,
-    "gpu_update_seconds": 0.9598710088771705,
-    "producer_batches_per_second": 9.999370992701923,
-    "gpu_batches_per_second": 1.041806649801593,
-    "producer_headroom": 9.59810632290191,
+    "producer_batch_seconds": 0.08375000747662398,
+    "producer_p95_seconds": 0.21814831099982257,
+    "gpu_update_seconds": 0.9616247472555196,
+    "producer_batches_per_second": 11.940297441515055,
+    "gpu_batches_per_second": 1.039906681742544,
+    "producer_headroom": 11.482085509352641,
     "sustainable": true,
-    "buffer_seconds": 60.47187355926174,
+    "buffer_seconds": 60.582359077097735,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 481629.1525466817,
-    "upload_bytes_per_second": 5601185.343684633,
+    "generation_bytes_per_second": 481375.57055169076,
+    "upload_bytes_per_second": 5558394.964128252,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.34640438366259,
+    "estimated_full_shards_per_hour": 103.29199159062426,
     "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": 1.1673664749996533,
-      "bytes": 6863070,
-      "pages": 38,
-      "shards": 3
-    },
     {
       "seconds": 2.7079190219992597,
       "bytes": 5909001,
@@ -562,6 +556,12 @@
       "bytes": 6092917,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9817484880004486,
+      "bytes": 2589304,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -162,3 +162,4 @@
 {"step": 413300, "loss": 0.5443356037139893, "lr": 0.0003, "grad_norm": 0.13792302804523293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60496215030, "tokens_per_second": 151755.15753279935, "data_wait_s": 1.3458000466926023e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 413400, "loss": 0.5988722443580627, "lr": 0.0003, "grad_norm": 0.1546253345086382, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60510814894, "tokens_per_second": 150968.00532214882, "data_wait_s": 2.282399873365648e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 413500, "loss": 0.6189818382263184, "lr": 0.0003, "grad_norm": 0.14126129704596418, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60525458209, "tokens_per_second": 151907.4258090089, "data_wait_s": 1.3625001884065568e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 413600, "loss": 0.6536849141120911, "lr": 0.0003, "grad_norm": 0.1670448951546124, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 60540127688, "tokens_per_second": 151910.38944963424, "data_wait_s": 1.9122999219689518e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ac91506dc3e2b04ad43cab82559ae4b999aa5bdc584e57bf234fa051546219ea
-size 349866
+oid sha256:d436a1dbebcd14dffea1fc309933dc7a30b0b8b7178faad7a320cb6dc01d4849
+size 351984