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": 75088.73432612585,
-      "required_pages_per_second_upper_estimate": 0.005194812870549214
+      "required_token_positions_per_second": 77422.49903762172,
+      "required_pages_per_second_upper_estimate": 0.005356268128370681
     },
     {
       "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": 9386.091790765731,
-      "required_pages_per_second_upper_estimate": 0.003929760628908676
+      "required_token_positions_per_second": 9677.812379702715,
+      "required_pages_per_second_upper_estimate": 0.00405189794767797
     },
     {
       "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": 18772.183581531463,
-      "required_pages_per_second_upper_estimate": 0.0039097507842609
+      "required_token_positions_per_second": 19355.62475940543,
+      "required_pages_per_second_upper_estimate": 0.004031266195233497
     },
     {
       "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": 28158.275372297197,
-      "required_pages_per_second_upper_estimate": 0.001688045613372226
+      "required_token_positions_per_second": 29033.437139108144,
+      "required_pages_per_second_upper_estimate": 0.0017405102249978984
     },
     {
       "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": 28158.275372297197,
-      "required_pages_per_second_upper_estimate": 0.0021891820550321616
+      "required_token_positions_per_second": 29033.437139108144,
+      "required_pages_per_second_upper_estimate": 0.002257222032971921
     },
     {
       "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": 28158.275372297197,
-      "required_pages_per_second_upper_estimate": 0.3611841224753043
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 29033.437139108144,
+      "required_pages_per_second_upper_estimate": 0.37240975794446124
     }
   ],
   "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": 62,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08895109720288019,
-    "producer_p95_seconds": 0.3315481230019941,
-    "gpu_update_seconds": 0.9818783158573974,
-    "producer_batches_per_second": 11.24213226644292,
-    "gpu_batches_per_second": 1.0184561404910732,
-    "producer_headroom": 11.03840589642108,
+    "producer_batch_seconds": 0.11199343928120697,
+    "producer_p95_seconds": 0.33969857600459363,
+    "gpu_update_seconds": 0.9522813254086973,
+    "producer_batches_per_second": 8.929094475695816,
+    "gpu_batches_per_second": 1.0501098502281592,
+    "producer_headroom": 8.503009922015089,
     "sustainable": true,
-    "buffer_seconds": 60.87645558315864,
+    "buffer_seconds": 60.946004826156624,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 328870.0340205267,
-    "upload_bytes_per_second": 4915657.062300158,
+    "generation_bytes_per_second": 342644.7653471261,
+    "upload_bytes_per_second": 4927915.1793318195,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.56785359823084,
+    "estimated_full_shards_per_hour": 73.5235902815851,
     "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": 1.0108026320012868,
-      "bytes": 815856,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 0.99785261299985,
       "bytes": 326991,
@@ -562,6 +556,12 @@
       "bytes": 5793744,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.8524275599993416,
+      "bytes": 1090009,
+      "pages": 35,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -356,3 +356,4 @@
 {"step": 592700, "loss": 0.5813198685646057, "lr": 0.0003, "grad_norm": 0.14987911509467272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86753634337, "tokens_per_second": 149733.81975582172, "data_wait_s": 8.442999387625605e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 592800, "loss": 0.617486298084259, "lr": 0.0003, "grad_norm": 0.1659861888705255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86768169057, "tokens_per_second": 150884.67283277473, "data_wait_s": 1.1156000255141407e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 592900, "loss": 0.5596344470977783, "lr": 0.0003, "grad_norm": 0.14808720066284142, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86782788006, "tokens_per_second": 151743.887041233, "data_wait_s": 1.5892001101747155e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 593000, "loss": 0.5732935070991516, "lr": 0.0003, "grad_norm": 0.1371412965224625, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 86797405638, "tokens_per_second": 141806.3136791834, "data_wait_s": 1.210100163007155e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2c8863a50572883df140c95410c1e791057792c7737c62bbd331acadec969d23
-size 763890
+oid sha256:9a14d112f9b516e1df12aa92289250068f140ff5a13ba3e3178ef122e5e657bf
+size 766008