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": 75034.7584513337,
-      "required_pages_per_second_upper_estimate": 0.00519107869428981
+      "required_token_positions_per_second": 76285.19863579908,
+      "required_pages_per_second_upper_estimate": 0.00527758704767209
     },
     {
       "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": 9379.344806416713,
-      "required_pages_per_second_upper_estimate": 0.003926935807454797
+      "required_token_positions_per_second": 9535.649829474885,
+      "required_pages_per_second_upper_estimate": 0.0039923774565892075
     },
     {
       "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": 18758.689612833427,
-      "required_pages_per_second_upper_estimate": 0.003906940346440984
+      "required_token_positions_per_second": 19071.29965894977,
+      "required_pages_per_second_upper_estimate": 0.003972048774965712
     },
     {
       "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": 28138.03441925014,
-      "required_pages_per_second_upper_estimate": 0.0016868321991432007
+      "required_token_positions_per_second": 28606.949488424656,
+      "required_pages_per_second_upper_estimate": 0.0017149429415483598
     },
     {
       "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": 28138.03441925014,
-      "required_pages_per_second_upper_estimate": 0.002187608409963296
+      "required_token_positions_per_second": 28606.949488424656,
+      "required_pages_per_second_upper_estimate": 0.0022240644940522015
     },
     {
       "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": 28138.03441925014,
-      "required_pages_per_second_upper_estimate": 0.36092449326265874
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28606.949488424656,
+      "required_pages_per_second_upper_estimate": 0.3669392322882551
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.12054006738289047,
-    "producer_p95_seconds": 0.23207776100025512,
-    "gpu_update_seconds": 0.9825846250683775,
-    "producer_batches_per_second": 8.295996689827142,
-    "gpu_batches_per_second": 1.0177240458351469,
-    "producer_headroom": 8.151518797042304,
+    "producer_batch_seconds": 0.08513911323430534,
+    "producer_p95_seconds": 0.21878025399928447,
+    "gpu_update_seconds": 0.9664784429806932,
+    "producer_batches_per_second": 11.745482916271051,
+    "gpu_batches_per_second": 1.0346842262885076,
+    "producer_headroom": 11.351756040973978,
     "sustainable": true,
-    "buffer_seconds": 60.92024675423941,
+    "buffer_seconds": 60.88814190778367,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 77,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 357457.9618999792,
+    "generation_bytes_per_second": 343672.3645993195,
     "upload_bytes_per_second": 5039535.789651298,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.70215742826016,
+    "estimated_full_shards_per_hour": 73.7440891598195,
     "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-v1/1790868774355361490-de221e1d/train.jsonl

@@ -127,3 +127,4 @@
 {"step": 452900, "loss": 0.5924435257911682, "lr": 0.0003, "grad_norm": 0.19952839026130523, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66290400713, "tokens_per_second": 150704.25808922146, "data_wait_s": 1.7930999092641287e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 453000, "loss": 0.5749834775924683, "lr": 0.0003, "grad_norm": 0.16922028082554208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66304918494, "tokens_per_second": 151102.69570079457, "data_wait_s": 1.3491000572685152e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 453100, "loss": 0.6096059679985046, "lr": 0.0003, "grad_norm": 0.1574331178666429, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66319640120, "tokens_per_second": 153029.26705929264, "data_wait_s": 1.4740000551682897e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 453200, "loss": 0.5835887789726257, "lr": 0.0003, "grad_norm": 0.1640516525907566, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66334325552, "tokens_per_second": 153203.19479277902, "data_wait_s": 6.297999789239839e-06, "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:a577b462c54f888d0966eec98b09a6ab5c819cd046d674c0034292e9ac0c5ced
-size 274953
+oid sha256:86f2fe87fb89d6e4b391530469d31af91077a6e605ed5e461a324b11bcbe72e9
+size 277071