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": 75847.96183570247,
-      "required_pages_per_second_upper_estimate": 0.005247337991312256
+      "required_token_positions_per_second": 75034.7584513337,
+      "required_pages_per_second_upper_estimate": 0.00519107869428981
     },
     {
       "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": 9480.99522946281,
-      "required_pages_per_second_upper_estimate": 0.003969494716881983
+      "required_token_positions_per_second": 9379.344806416713,
+      "required_pages_per_second_upper_estimate": 0.003926935807454797
     },
     {
       "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": 18961.99045892562,
-      "required_pages_per_second_upper_estimate": 0.003949282551278138
+      "required_token_positions_per_second": 18758.689612833427,
+      "required_pages_per_second_upper_estimate": 0.003906940346440984
     },
     {
       "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": 28442.985688388428,
-      "required_pages_per_second_upper_estimate": 0.0017051135620944143
+      "required_token_positions_per_second": 28138.03441925014,
+      "required_pages_per_second_upper_estimate": 0.0016868321991432007
     },
     {
       "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": 28442.985688388428,
-      "required_pages_per_second_upper_estimate": 0.0022113170297999225
+      "required_token_positions_per_second": 28138.03441925014,
+      "required_pages_per_second_upper_estimate": 0.002187608409963296
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28442.985688388428,
-      "required_pages_per_second_upper_estimate": 0.3648360807120025
+      "required_token_positions_per_second": 28138.03441925014,
+      "required_pages_per_second_upper_estimate": 0.36092449326265874
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.18209748068747444,
-    "producer_p95_seconds": 0.2345397970002523,
-    "gpu_update_seconds": 0.9720498509861792,
-    "producer_batches_per_second": 5.491564167854986,
-    "gpu_batches_per_second": 1.028753822641364,
-    "producer_headroom": 5.338074131044481,
+    "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,
     "sustainable": true,
-    "buffer_seconds": 60.267090761143116,
+    "buffer_seconds": 60.92024675423941,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 77,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 343242.8810743897,
-    "upload_bytes_per_second": 5091445.966110806,
+    "generation_bytes_per_second": 357457.9618999792,
+    "upload_bytes_per_second": 5039535.789651298,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 73.65193199323433,
+    "estimated_full_shards_per_hour": 76.70215742826016,
     "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": 1.07210781699996,
-      "bytes": 6934469,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0243737840000904,
       "bytes": 7482664,
@@ -562,6 +556,12 @@
       "bytes": 11191807,
       "pages": 40,
       "shards": 3
+    },
+    {
+      "seconds": 1.0167957749999914,
+      "bytes": 2251564,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -126,3 +126,4 @@
 {"step": 452800, "loss": 0.5833898186683655, "lr": 0.0003, "grad_norm": 0.1638339779408624, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66275983640, "tokens_per_second": 151077.60845432733, "data_wait_s": 1.3533999663195573e-05, "gpu_peak_gb": 89.37899398803711}
 {"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}

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 sha256:54fdf94e04598aa5fc41187bb38a8a5dc3f636222c3f8d54b3ede055353058c6
-size 272835
+oid sha256:a577b462c54f888d0966eec98b09a6ab5c819cd046d674c0034292e9ac0c5ced
+size 274953