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": 75595.39108225425,
-      "required_pages_per_second_upper_estimate": 0.005229864560543818
+      "required_token_positions_per_second": 76473.24634729626,
+      "required_pages_per_second_upper_estimate": 0.005290596624684269
     },
     {
       "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": 9449.423885281782,
-      "required_pages_per_second_upper_estimate": 0.003956276454358023
+      "required_token_positions_per_second": 9559.155793412032,
+      "required_pages_per_second_upper_estimate": 0.004002218912829346
     },
     {
       "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": 18898.847770563563,
-      "required_pages_per_second_upper_estimate": 0.003936131594476993
+      "required_token_positions_per_second": 19118.311586824064,
+      "required_pages_per_second_upper_estimate": 0.003981840119754017
     },
     {
       "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": 28348.271655845343,
-      "required_pages_per_second_upper_estimate": 0.0016994356004634103
+      "required_token_positions_per_second": 28677.467380236096,
+      "required_pages_per_second_upper_estimate": 0.0017191703814878692
     },
     {
       "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": 28348.271655845343,
-      "required_pages_per_second_upper_estimate": 0.0022039534303726987
+      "required_token_positions_per_second": 28677.467380236096,
+      "required_pages_per_second_upper_estimate": 0.002229546949965115
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28348.271655845343,
-      "required_pages_per_second_upper_estimate": 0.3636211907985447
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28677.467380236096,
+      "required_pages_per_second_upper_estimate": 0.3678437600881992
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.09058547213263068,
-    "producer_p95_seconds": 0.26583486199888284,
-    "gpu_update_seconds": 0.9752975537857543,
-    "producer_batches_per_second": 11.039297764390414,
-    "gpu_batches_per_second": 1.0253281125522766,
-    "producer_headroom": 10.766600105122517,
+    "producer_batch_seconds": 0.15116301550796152,
+    "producer_p95_seconds": 0.2661653950017353,
+    "gpu_update_seconds": 0.9641018725054646,
+    "producer_batches_per_second": 6.61537477695615,
+    "gpu_batches_per_second": 1.0372347866115486,
+    "producer_headroom": 6.377895209788845,
     "sustainable": true,
-    "buffer_seconds": 60.46844833471677,
+    "buffer_seconds": 60.73841796784427,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 454749.41652648116,
+    "generation_bytes_per_second": 455996.8377349973,
     "upload_bytes_per_second": 5165004.004546632,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 97.57863876195742,
+    "estimated_full_shards_per_hour": 97.84630631482543,
     "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-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -209,3 +209,4 @@
 {"step": 740900, "loss": 0.5532585382461548, "lr": 5e-05, "grad_norm": 0.22479074977986213, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108447557034, "tokens_per_second": 149233.26902220756, "data_wait_s": 1.9133000023430213e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 741000, "loss": 0.4990701377391815, "lr": 5e-05, "grad_norm": 0.22058791630189842, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108462150243, "tokens_per_second": 150116.16704524864, "data_wait_s": 1.9185001292498782e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 741100, "loss": 0.5006605386734009, "lr": 5e-05, "grad_norm": 0.22714152903226234, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108476746284, "tokens_per_second": 150221.69880708976, "data_wait_s": 2.152199886040762e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 741200, "loss": 0.5441394448280334, "lr": 5e-05, "grad_norm": 0.230356178227827, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108491474951, "tokens_per_second": 151107.16170434657, "data_wait_s": 1.8865001038648188e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cce171c030a18bfa616b03c877a0b8a2166dee5c4102972ffe088530e6714d0f
-size 450195
+oid sha256:669e488ef624b1475d999ccad84f873ab14d3f581d16158f7c11b0bff3b54b55
+size 452313