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": 76200.71080982039,
-      "required_pages_per_second_upper_estimate": 0.005271741983832119
+      "required_token_positions_per_second": 77429.58956040429,
+      "required_pages_per_second_upper_estimate": 0.005356758667189059
     },
     {
       "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": 9525.088851227549,
-      "required_pages_per_second_upper_estimate": 0.003987955795535352
+      "required_token_positions_per_second": 9678.698695050536,
+      "required_pages_per_second_upper_estimate": 0.004052269029405734
     },
     {
       "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": 19050.177702455097,
-      "required_pages_per_second_upper_estimate": 0.003967649628451322
+      "required_token_positions_per_second": 19357.397390101072,
+      "required_pages_per_second_upper_estimate": 0.004031635387459985
     },
     {
       "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": 28575.266553682643,
-      "required_pages_per_second_upper_estimate": 0.0017130435979876163
+      "required_token_positions_per_second": 29036.096085151607,
+      "required_pages_per_second_upper_estimate": 0.0017406696247532286
     },
     {
       "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": 28575.266553682643,
-      "required_pages_per_second_upper_estimate": 0.002221601285234512
+      "required_token_positions_per_second": 29036.096085151607,
+      "required_pages_per_second_upper_estimate": 0.0022574287543313323
     },
     {
       "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": 28575.266553682643,
-      "required_pages_per_second_upper_estimate": 0.36653283761986943
+      "required_token_positions_per_second": 29036.096085151607,
+      "required_pages_per_second_upper_estimate": 0.372443864049353
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.1515560201094388,
-    "producer_p95_seconds": 0.27276480100044864,
-    "gpu_update_seconds": 0.9675500296054756,
-    "producer_batches_per_second": 6.598220244091252,
-    "gpu_batches_per_second": 1.0335382868085448,
-    "producer_headroom": 6.384108192513939,
+    "producer_batch_seconds": 0.20260347635925768,
+    "producer_p95_seconds": 0.2658181139995577,
+    "gpu_update_seconds": 0.9521941213763427,
+    "producer_batches_per_second": 4.935749464766311,
+    "gpu_batches_per_second": 1.0502060215983655,
+    "producer_headroom": 4.699791624936911,
     "sustainable": true,
-    "buffer_seconds": 60.955651865144965,
+    "buffer_seconds": 60.940423768085935,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 516723.0175051098,
-    "upload_bytes_per_second": 4777160.797231375,
+    "generation_bytes_per_second": 513330.24765047425,
+    "upload_bytes_per_second": 4818821.457835323,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 110.87673086037607,
+    "estimated_full_shards_per_hour": 110.14872142921134,
     "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": 1.2706896670024435,
-      "bytes": 7639609,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 1.257235208999191,
       "bytes": 5936765,
@@ -562,6 +556,12 @@
       "bytes": 6151647,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.3665102609993482,
+      "bytes": 11866902,
+      "pages": 32,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -242,3 +242,4 @@
 {"step": 464400, "loss": 0.5809440612792969, "lr": 0.0003, "grad_norm": 0.17934050200999524, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67973966158, "tokens_per_second": 150585.80848647223, "data_wait_s": 1.7591999494470656e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 464500, "loss": 0.5573697686195374, "lr": 0.0003, "grad_norm": 0.16440483884860438, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 67988658543, "tokens_per_second": 152803.34575908878, "data_wait_s": 9.708001016406342e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 464600, "loss": 0.6125659942626953, "lr": 0.0003, "grad_norm": 0.1611607892049086, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68003346812, "tokens_per_second": 152839.153010274, "data_wait_s": 1.1713000276358798e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 464700, "loss": 0.5485379695892334, "lr": 0.0003, "grad_norm": 0.1977374657831521, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68017987997, "tokens_per_second": 152362.4185223103, "data_wait_s": 1.6730002244003117e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d49f2c66e0d3a9bfdc47adaebc8b80687ee1771f8b66da230d836be2fa787816
-size 520089
+oid sha256:f5e413d441491a3c74f592271894de11e8db70fc2048cea387f3e94a4df938c0
+size 522207