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": 75980.90530261012,
-      "required_pages_per_second_upper_estimate": 0.006585832299786333
+      "required_token_positions_per_second": 74492.85061151587,
+      "required_pages_per_second_upper_estimate": 0.006456851490602418
     },
     {
       "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": 9497.613162826265,
-      "required_pages_per_second_upper_estimate": 0.0037887398926225724
+      "required_token_positions_per_second": 9311.606326439483,
+      "required_pages_per_second_upper_estimate": 0.0037145389845378504
     },
     {
       "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": 18995.22632565253,
-      "required_pages_per_second_upper_estimate": 0.0043942113058743
+      "required_token_positions_per_second": 18623.212652878967,
+      "required_pages_per_second_upper_estimate": 0.004308152489895162
     },
     {
       "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": 28492.839488478792,
-      "required_pages_per_second_upper_estimate": 0.0017350644344727873
+      "required_token_positions_per_second": 27934.818979318447,
+      "required_pages_per_second_upper_estimate": 0.0017010839131722686
     },
     {
       "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": 28492.839488478792,
-      "required_pages_per_second_upper_estimate": 0.0021934576570509126
+      "required_token_positions_per_second": 27934.818979318447,
+      "required_pages_per_second_upper_estimate": 0.0021504996935561146
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28492.839488478792,
-      "required_pages_per_second_upper_estimate": 0.353785707047429
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 27934.818979318447,
+      "required_pages_per_second_upper_estimate": 0.3468569598981642
     }
   ],
   "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": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 717798,
-    "producer_batch_seconds": 0.13911894355476306,
-    "producer_p95_seconds": 0.38204476299870294,
-    "gpu_update_seconds": 0.9703490594954425,
-    "producer_batches_per_second": 7.188093687660573,
-    "gpu_batches_per_second": 1.0305569838136137,
-    "producer_headroom": 6.974959949386564,
+    "producer_batch_seconds": 0.09221069406254401,
+    "producer_p95_seconds": 0.3569645669995225,
+    "gpu_update_seconds": 0.9897325635247254,
+    "producer_batches_per_second": 10.844729130026145,
+    "gpu_batches_per_second": 1.0103739503515063,
+    "producer_headroom": 10.733381562592042,
     "sustainable": true,
-    "buffer_seconds": 60.16164168871743,
+    "buffer_seconds": 60.37368637500825,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 86,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311080.7886935105,
-    "upload_bytes_per_second": 5301976.808918721,
+    "generation_bytes_per_second": 290145.6995019947,
+    "upload_bytes_per_second": 5260029.988829781,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.75069566349016,
+    "estimated_full_shards_per_hour": 62.25851286692506,
     "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": 1.218213135998667,
-      "bytes": 11903355,
-      "pages": 42,
-      "shards": 4
-    },
     {
       "seconds": 1.0320353019997128,
       "bytes": 1021485,
@@ -562,6 +556,12 @@
       "bytes": 6196482,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.9420911959969089,
+      "bytes": 7110283,
+      "pages": 25,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790782192342215557-281f6532/train.jsonl

@@ -337,3 +337,4 @@
 {"step": 388700, "loss": 0.6117727477103472, "lr": 0.0003, "grad_norm": 0.16343420983775175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56895205001, "tokens_per_second": 149016.2557481882, "data_wait_s": 1.4340999769046903e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 388800, "loss": 0.5917207412421703, "lr": 0.0003, "grad_norm": 0.14948196008270812, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56909767722, "tokens_per_second": 149126.38797229237, "data_wait_s": 1.5354999050032347e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 388900, "loss": 0.5778171997517347, "lr": 0.0003, "grad_norm": 0.16686476276738277, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56924420895, "tokens_per_second": 146948.23268215667, "data_wait_s": 1.3607001164928079e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 389000, "loss": 0.5543361138552427, "lr": 0.0003, "grad_norm": 0.13831569021399448, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 56939050728, "tokens_per_second": 126040.32964440901, "data_wait_s": 1.6482998034916818e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790782192.062ccfe6-664f-43fe-9e0e-62f43f0c4870-64tpq.429.0.1790782192342215557-281f6532

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7933076dd2dd9ecdab19e0fafc8fe9b8ee07cb6c63ddf46871641f695a122178
-size 722865
+oid sha256:819ab3508fd337c89397c58ffd87e74f2795484d1582c370feeffd8e18e0e688
+size 724983