Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75747.29837479378,
-      "required_pages_per_second_upper_estimate": 0.006565583843354559
+      "required_token_positions_per_second": 75996.61361005671,
+      "required_pages_per_second_upper_estimate": 0.006587193856063462
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9468.412296849223,
-      "required_pages_per_second_upper_estimate": 0.0037770912305924776
+      "required_token_positions_per_second": 9499.576701257089,
+      "required_pages_per_second_upper_estimate": 0.0037895231774601437
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18936.824593698446,
-      "required_pages_per_second_upper_estimate": 0.0043807010930221
+      "required_token_positions_per_second": 18999.153402514177,
+      "required_pages_per_second_upper_estimate": 0.00439511976599205
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28405.236890547665,
-      "required_pages_per_second_upper_estimate": 0.001729729895874092
+      "required_token_positions_per_second": 28498.730103771264,
+      "required_pages_per_second_upper_estimate": 0.001735423142038429
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28405.236890547665,
-      "required_pages_per_second_upper_estimate": 0.002186713766562663
+      "required_token_positions_per_second": 28498.730103771264,
+      "required_pages_per_second_upper_estimate": 0.002193911132922394
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28405.236890547665,
-      "required_pages_per_second_upper_estimate": 0.35269797596815955
+      "required_token_positions_per_second": 28498.730103771264,
+      "required_pages_per_second_upper_estimate": 0.35385884877474033
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.05977866765633166,
-    "producer_p95_seconds": 0.19283926999924006,
-    "gpu_update_seconds": 0.9733416449415477,
-    "producer_batches_per_second": 16.728375509287243,
-    "gpu_batches_per_second": 1.0273884870713132,
-    "producer_headroom": 16.282424535409547,
+    "producer_batch_seconds": 0.26845251605465137,
+    "producer_p95_seconds": 0.22487620799984143,
+    "gpu_update_seconds": 0.9701484908038521,
+    "producer_batches_per_second": 3.7250535576892143,
+    "gpu_batches_per_second": 1.0307700413690417,
+    "producer_headroom": 3.6138550871557116,
     "sustainable": true,
-    "buffer_seconds": 60.34718198637596,
+    "buffer_seconds": 60.149206429838834,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 80,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 492046.17694831773,
-    "upload_bytes_per_second": 5385503.514001928,
+    "generation_bytes_per_second": 479385.33446691424,
+    "upload_bytes_per_second": 5231281.311811079,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.58165532433652,
+    "estimated_full_shards_per_hour": 102.8649332571561,
     "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": 1.4588776000000507,
-      "bytes": 16176314,
-      "pages": 44,
-      "shards": 4
-    },
     {
       "seconds": 0.8323371429996769,
       "bytes": 2088001,
@@ -562,6 +556,12 @@
       "bytes": 5875792,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 0.9227722110008472,
+      "bytes": 1091891,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -126,3 +126,4 @@
 {"step": 167600, "loss": 0.57669323310256, "lr": 0.0003, "grad_norm": 0.12682190382748396, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24528941816, "tokens_per_second": 149596.6484980393, "data_wait_s": 1.9020999388885684e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 167700, "loss": 0.6204749215394258, "lr": 0.0003, "grad_norm": 0.14254141648881558, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24543636953, "tokens_per_second": 149899.0740025255, "data_wait_s": 1.0965999535983428e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 167800, "loss": 0.6374548450112343, "lr": 0.0003, "grad_norm": 0.16923725250118113, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24558282554, "tokens_per_second": 149795.56277722394, "data_wait_s": 1.1805999747593887e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 167900, "loss": 0.5849163699895144, "lr": 0.0003, "grad_norm": 0.1398667099829425, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 24572900308, "tokens_per_second": 149857.7722695745, "data_wait_s": 1.554099981149193e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ee6c99e054ac539db5bccfd127818b446f1ef9a37639a7f3f270880e9f1e9c04
-size 272835
+oid sha256:32c161e74635dfa7deb0f63551a1ba2d158c2639de06dbad621f36a6cc6d843d
+size 274953