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": 76356.00858976411,
-      "required_pages_per_second_upper_estimate": 0.00528248584458912
+      "required_token_positions_per_second": 76504.21342400821,
+      "required_pages_per_second_upper_estimate": 0.00529273899890472
     },
     {
       "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": 9544.501073720514,
-      "required_pages_per_second_upper_estimate": 0.003996083287719854
+      "required_token_positions_per_second": 9563.026678001026,
+      "required_pages_per_second_upper_estimate": 0.004003839571373496
     },
     {
       "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": 19089.002147441028,
-      "required_pages_per_second_upper_estimate": 0.003975735736472426
+      "required_token_positions_per_second": 19126.053356002052,
+      "required_pages_per_second_upper_estimate": 0.003983452526109609
     },
     {
       "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": 28633.503221161543,
-      "required_pages_per_second_upper_estimate": 0.0017165347972807339
+      "required_token_positions_per_second": 28689.08003400308,
+      "required_pages_per_second_upper_estimate": 0.0017198665423497044
     },
     {
       "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": 28633.503221161543,
-      "required_pages_per_second_upper_estimate": 0.0022261289299749677
+      "required_token_positions_per_second": 28689.08003400308,
+      "required_pages_per_second_upper_estimate": 0.0022304497827168326
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 28633.503221161543,
-      "required_pages_per_second_upper_estimate": 0.367279835060627
+      "required_token_positions_per_second": 28689.08003400308,
+      "required_pages_per_second_upper_estimate": 0.3679927147420259
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.07203993300774414,
-    "producer_p95_seconds": 0.18900574699910067,
-    "gpu_update_seconds": 0.9655821638885874,
-    "producer_batches_per_second": 13.881190032374157,
-    "gpu_batches_per_second": 1.035644647756132,
-    "producer_headroom": 13.40342950880853,
+    "producer_batch_seconds": 0.34449462380457874,
+    "producer_p95_seconds": 0.18574844099930488,
+    "gpu_update_seconds": 0.9637116271149454,
+    "producer_batches_per_second": 2.9028029202779937,
+    "gpu_batches_per_second": 1.0376548044705975,
+    "producer_headroom": 2.7974649254951203,
     "sustainable": true,
-    "buffer_seconds": 60.83167632498101,
+    "buffer_seconds": 60.71383250824156,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 89,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 485026.75635320565,
-    "upload_bytes_per_second": 5029592.1244579265,
+    "generation_bytes_per_second": 429648.45356901037,
+    "upload_bytes_per_second": 5020266.484003271,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 104.07545106837394,
+    "estimated_full_shards_per_hour": 92.19255643179639,
     "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": 1.117540433999693,
-      "bytes": 7055686,
-      "pages": 40,
-      "shards": 2
-    },
     {
       "seconds": 1.0181373959999291,
       "bytes": 10411775,
@@ -562,6 +556,12 @@
       "bytes": 7684006,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1074004389993206,
+      "bytes": 6213956,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -93,3 +93,4 @@
 {"step": 406400, "loss": 0.5909572243690491, "lr": 0.0003, "grad_norm": 0.1529850649040701, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59486498541, "tokens_per_second": 153533.83831338357, "data_wait_s": 1.0314999599358998e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 406500, "loss": 0.5848004221916199, "lr": 0.0003, "grad_norm": 0.16868212642409153, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59501288633, "tokens_per_second": 154018.16408604814, "data_wait_s": 1.0071000360767357e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 406600, "loss": 0.6084566116333008, "lr": 0.0003, "grad_norm": 0.16289168414628516, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59515798258, "tokens_per_second": 151487.60826619124, "data_wait_s": 9.319001037511043e-06, "gpu_peak_gb": 89.37861251831055}
+{"step": 406700, "loss": 0.6016278862953186, "lr": 0.0003, "grad_norm": 0.16093596745431255, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 59530322116, "tokens_per_second": 150622.5098614456, "data_wait_s": 1.6726000467315316e-05, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e9de710a9587d8c2fac185892900c656c8b8bcc2ad905fd75d2aa3bd17bcae24
-size 202941
+oid sha256:f66e63a80109e2467966a156f864654e2f7d4da339bad6a470cdf9ef6007fd3c
+size 205059