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": 76930.89631519059,
-      "required_pages_per_second_upper_estimate": 0.005322257911357426
+      "required_token_positions_per_second": 77538.42863327857,
+      "required_pages_per_second_upper_estimate": 0.005364288406791945
     },
     {
       "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": 9616.362039398824,
-      "required_pages_per_second_upper_estimate": 0.004026169973421757
+      "required_token_positions_per_second": 9692.303579159821,
+      "required_pages_per_second_upper_estimate": 0.004057965110280005
     },
     {
       "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": 19232.724078797648,
-      "required_pages_per_second_upper_estimate": 0.004005669224571799
+      "required_token_positions_per_second": 19384.607158319643,
+      "required_pages_per_second_upper_estimate": 0.0040373024646100705
     },
     {
       "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": 28849.086118196472,
-      "required_pages_per_second_upper_estimate": 0.001729458662781954
+      "required_token_positions_per_second": 29076.910737479466,
+      "required_pages_per_second_upper_estimate": 0.001743116400839918
     },
     {
       "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": 28849.086118196472,
-      "required_pages_per_second_upper_estimate": 0.0022428895519704833
+      "required_token_positions_per_second": 29076.910737479466,
+      "required_pages_per_second_upper_estimate": 0.0022606019140251404
     },
     {
       "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": 28849.086118196472,
-      "required_pages_per_second_upper_estimate": 0.37004510098891075
+      "required_token_positions_per_second": 29076.910737479466,
+      "required_pages_per_second_upper_estimate": 0.37296739058605544
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09848728552336183,
-    "producer_p95_seconds": 0.28347011000005296,
-    "gpu_update_seconds": 0.9583665800269877,
-    "producer_batches_per_second": 10.15359489995075,
-    "gpu_batches_per_second": 1.0434420615667126,
-    "producer_headroom": 9.730866019245264,
+    "producer_batch_seconds": 0.08339229177346397,
+    "producer_p95_seconds": 0.2911834660008026,
+    "gpu_update_seconds": 0.9508575463748412,
+    "producer_batches_per_second": 11.991515987071208,
+    "gpu_batches_per_second": 1.0516822460025848,
+    "producer_headroom": 11.402223468781212,
     "sustainable": true,
-    "buffer_seconds": 60.377094541700224,
+    "buffer_seconds": 60.85488296798984,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 422925.4836099955,
-    "upload_bytes_per_second": 4672302.402095558,
+    "generation_bytes_per_second": 425873.88765351806,
+    "upload_bytes_per_second": 4703995.026162921,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 90.74996358132266,
+    "estimated_full_shards_per_hour": 91.38262245372921,
     "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": 1.6032572549993347,
-      "bytes": 5165320,
-      "pages": 25,
-      "shards": 2
-    },
     {
       "seconds": 1.8941422179996152,
       "bytes": 6019433,
@@ -562,6 +556,12 @@
       "bytes": 8480974,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.155970021001849,
+      "bytes": 5946411,
+      "pages": 30,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -315,3 +315,4 @@
 {"step": 916500, "loss": 0.5014942288398743, "lr": 3e-05, "grad_norm": 0.3275443557246973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134148926944, "tokens_per_second": 153661.86036729874, "data_wait_s": 1.1418000212870538e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 916600, "loss": 0.520254373550415, "lr": 3e-05, "grad_norm": 0.3151773875789409, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134163827405, "tokens_per_second": 155649.0062536393, "data_wait_s": 1.5124001947697252e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 916700, "loss": 0.5406615138053894, "lr": 3e-05, "grad_norm": 0.3208103976323103, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134178599235, "tokens_per_second": 153731.71901362273, "data_wait_s": 1.50500018207822e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 916800, "loss": 0.5456134676933289, "lr": 3e-05, "grad_norm": 0.3340009759926441, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134193216992, "tokens_per_second": 152473.58338664877, "data_wait_s": 1.3147000572644174e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9d6e48d2f365f4a0256930560d432c342b50d140da5e1ff08dd3e07a75a76df7
-size 676269
+oid sha256:096e05233b66649d1ac2666e4f17fe03ee1594369df606128f6f37af45e1acbb
+size 678387