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": 76574.82674765578,
-      "required_pages_per_second_upper_estimate": 0.005297624192480144
+      "required_token_positions_per_second": 76120.39592065166,
+      "required_pages_per_second_upper_estimate": 0.005266185613443208
     },
     {
       "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": 9571.853343456973,
-      "required_pages_per_second_upper_estimate": 0.004007535111878166
+      "required_token_positions_per_second": 9515.049490081457,
+      "required_pages_per_second_upper_estimate": 0.003983752524669184
     },
     {
       "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": 19143.706686913945,
-      "required_pages_per_second_upper_estimate": 0.003987129249388914
+      "required_token_positions_per_second": 19030.098980162915,
+      "required_pages_per_second_upper_estimate": 0.003963467760109376
     },
     {
       "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": 28715.560030370918,
-      "required_pages_per_second_upper_estimate": 0.0017214539776993429
+      "required_token_positions_per_second": 28545.148470244374,
+      "required_pages_per_second_upper_estimate": 0.0017112380648731426
     },
     {
       "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": 28715.560030370918,
-      "required_pages_per_second_upper_estimate": 0.002232508485961237
+      "required_token_positions_per_second": 28545.148470244374,
+      "required_pages_per_second_upper_estimate": 0.002219259736722635
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "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": 28715.560030370918,
-      "required_pages_per_second_upper_estimate": 0.3683323717034276
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28545.148470244374,
+      "required_pages_per_second_upper_estimate": 0.3661465151838018
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07278603475795364,
-    "producer_p95_seconds": 0.18649154199920304,
-    "gpu_update_seconds": 0.9628229423614996,
-    "producer_batches_per_second": 13.738899272716951,
-    "gpu_batches_per_second": 1.0386125589688555,
-    "producer_headroom": 13.228127422565601,
+    "producer_batch_seconds": 0.08627264346880281,
+    "producer_p95_seconds": 0.1925725010005408,
+    "gpu_update_seconds": 0.9685708949393076,
+    "producer_batches_per_second": 11.591159836914139,
+    "gpu_batches_per_second": 1.0324489464064082,
+    "producer_headroom": 11.226860056624485,
     "sustainable": true,
-    "buffer_seconds": 60.657845368774474,
+    "buffer_seconds": 60.05139548623707,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 344502.0130549771,
-    "upload_bytes_per_second": 5721633.25147594,
+    "generation_bytes_per_second": 351808.47272401873,
+    "upload_bytes_per_second": 5680545.964801748,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.92211240517602,
+    "estimated_full_shards_per_hour": 75.48990856447622,
     "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": 1.4614262610000424,
-      "bytes": 6002577,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.2785650519999763,
       "bytes": 7789202,
@@ -562,6 +556,12 @@
       "bytes": 10529948,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 1.3451746110004024,
+      "bytes": 2254637,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791039222312944374-63c10d34/train.jsonl

@@ -81,3 +81,4 @@
 {"step": 612500, "loss": 0.5020858645439148, "lr": 0.0001, "grad_norm": 0.1588659803275934, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89652207771, "tokens_per_second": 151629.30302031912, "data_wait_s": 8.142998922266997e-06, "gpu_peak_gb": 89.37861251831055}
 {"step": 612600, "loss": 0.5658448934555054, "lr": 0.0001, "grad_norm": 0.14758167540133071, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89666718497, "tokens_per_second": 148572.6852950895, "data_wait_s": 1.8473433170001954, "gpu_peak_gb": 89.37861251831055}
 {"step": 612700, "loss": 0.5631892681121826, "lr": 0.0001, "grad_norm": 0.16110160584369795, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89681284322, "tokens_per_second": 151059.65530886315, "data_wait_s": 1.7604999811737798e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 612800, "loss": 0.566730260848999, "lr": 0.0001, "grad_norm": 0.1414385277889316, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 89695789065, "tokens_per_second": 150874.49695636056, "data_wait_s": 1.4975999874877743e-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 sha256:3b152c2fa2e1219f5e7e4be21570372788d80d80151428107263c08debfe606c
-size 177525
+oid sha256:1a6e9d02584e07716750c2a1f6dd3d2c6058ecf7045a551dc3d9b4d8c555531c
+size 179643