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": 76824.96168441964,
-      "required_pages_per_second_upper_estimate": 0.005314929107798483
+      "required_token_positions_per_second": 75674.70649172532,
+      "required_pages_per_second_upper_estimate": 0.005235351784608132
     },
     {
       "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": 9603.120210552455,
-      "required_pages_per_second_upper_estimate": 0.004020625896204613
+      "required_token_positions_per_second": 9459.338311465664,
+      "required_pages_per_second_upper_estimate": 0.003960427417564453
     },
     {
       "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": 19206.24042110491,
-      "required_pages_per_second_upper_estimate": 0.004000153377095422
+      "required_token_positions_per_second": 18918.67662293133,
+      "required_pages_per_second_upper_estimate": 0.003940261421503196
     },
     {
       "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": 28809.360631657368,
-      "required_pages_per_second_upper_estimate": 0.0017270771805212404
+      "required_token_positions_per_second": 28378.014934396993,
+      "required_pages_per_second_upper_estimate": 0.001701218664597225
     },
     {
       "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": 28809.360631657368,
-      "required_pages_per_second_upper_estimate": 0.002239801070126017
+      "required_token_positions_per_second": 28378.014934396993,
+      "required_pages_per_second_upper_estimate": 0.0022062658394532333
     },
     {
       "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": 28809.360631657368,
-      "required_pages_per_second_upper_estimate": 0.36953554510149134
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28378.014934396993,
+      "required_pages_per_second_upper_estimate": 0.3640027056399609
     }
   ],
   "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": 0.10259803892978425,
-    "producer_p95_seconds": 0.30950035299974843,
-    "gpu_update_seconds": 0.9596880803254898,
-    "producer_batches_per_second": 9.746774991326852,
-    "gpu_batches_per_second": 1.042005231179737,
-    "producer_headroom": 9.35386378079096,
+    "producer_batch_seconds": 0.13658434514823625,
+    "producer_p95_seconds": 0.33800842899654526,
+    "gpu_update_seconds": 0.974275334758805,
+    "producer_batches_per_second": 7.321483285033074,
+    "gpu_batches_per_second": 1.02640389664341,
+    "producer_headroom": 7.133140578456594,
     "sustainable": true,
-    "buffer_seconds": 60.460349060505855,
+    "buffer_seconds": 60.40507075504591,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 164,
+    "upload_pages_per_commit": 170,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 630872.0552711069,
-    "upload_bytes_per_second": 4786881.781326703,
+    "generation_bytes_per_second": 621008.8672862906,
+    "upload_bytes_per_second": 4778062.236941557,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.68522855567886,
+    "estimated_full_shards_per_hour": 66.62702328654069,
     "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": 1.060381152001355,
-      "bytes": 6181629,
-      "pages": 30,
-      "shards": 2
-    },
     {
       "seconds": 1.3323454749988741,
       "bytes": 6231984,
@@ -562,6 +556,12 @@
       "bytes": 8054475,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.189110659997823,
+      "bytes": 6011923,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -324,3 +324,4 @@
 {"step": 957400, "loss": 0.5411211848258972, "lr": 3e-05, "grad_norm": 0.3355946815973881, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140137228184, "tokens_per_second": 153261.5840807012, "data_wait_s": 9.708001016406342e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 957500, "loss": 0.5133512616157532, "lr": 3e-05, "grad_norm": 0.3413272440957866, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140151992194, "tokens_per_second": 153622.89164431932, "data_wait_s": 1.282700031879358e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 957600, "loss": 0.5616347789764404, "lr": 3e-05, "grad_norm": 0.3690775747272903, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140166772155, "tokens_per_second": 153383.5520363414, "data_wait_s": 1.1943997378693894e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 957700, "loss": 0.6012611389160156, "lr": 3e-05, "grad_norm": 0.3550671449172152, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 140181537401, "tokens_per_second": 154312.0595150318, "data_wait_s": 1.1293002899037674e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c302221fd67c366a62aa80d325dc2978962d39ca6d2c0b5d1b4dd06a6780bd71
-size 695331
+oid sha256:ca186dc9d203a64ed06a5c79fe45e3556640763e7269b932b032c875e12a4ace
+size 697449