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": 73612.05970681633,
-      "required_pages_per_second_upper_estimate": 0.006380506767327201
+      "required_token_positions_per_second": 74826.87545340792,
+      "required_pages_per_second_upper_estimate": 0.0064858039173193984
     },
     {
       "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": 9201.507463352042,
-      "required_pages_per_second_upper_estimate": 0.0036706189019275735
+      "required_token_positions_per_second": 9353.35943167599,
+      "required_pages_per_second_upper_estimate": 0.003731194922481247
     },
     {
       "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": 18403.014926704083,
-      "required_pages_per_second_upper_estimate": 0.0042572136212922235
+      "required_token_positions_per_second": 18706.71886335198,
+      "required_pages_per_second_upper_estimate": 0.004327470182028984
     },
     {
       "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": 27604.522390056125,
-      "required_pages_per_second_upper_estimate": 0.0016809705838184692
+      "required_token_positions_per_second": 28060.078295027968,
+      "required_pages_per_second_upper_estimate": 0.001708711548314139
     },
     {
       "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": 27604.522390056125,
-      "required_pages_per_second_upper_estimate": 0.0021250725477952232
+      "required_token_positions_per_second": 28060.078295027968,
+      "required_pages_per_second_upper_estimate": 0.002160142502419412
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27604.522390056125,
-      "required_pages_per_second_upper_estimate": 0.34275578169110005
+      "required_token_positions_per_second": 28060.078295027968,
+      "required_pages_per_second_upper_estimate": 0.3484122613833141
     }
   ],
   "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": 60,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.37117361967179363,
-    "producer_p95_seconds": 0.22055574499972863,
-    "gpu_update_seconds": 1.0015750176485407,
-    "producer_batches_per_second": 2.6941569847669657,
-    "gpu_batches_per_second": 0.9984274591310809,
-    "producer_headroom": 2.6984003295659127,
+    "producer_batch_seconds": 0.08451387303122715,
+    "producer_p95_seconds": 0.2664653850006289,
+    "gpu_update_seconds": 0.985314428181728,
+    "producer_batches_per_second": 11.832376912019031,
+    "gpu_batches_per_second": 1.0149044522217872,
+    "producer_headroom": 11.658611691096713,
     "sustainable": true,
-    "buffer_seconds": 60.09450105891244,
+    "buffer_seconds": 60.104180119085406,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 290147.71974332695,
-    "upload_bytes_per_second": 5223533.344609018,
+    "generation_bytes_per_second": 290587.8447057929,
+    "upload_bytes_per_second": 5266415.1016032,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 62.25894636368614,
+    "estimated_full_shards_per_hour": 62.35338693504658,
     "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": 1.2782940739998594,
-      "bytes": 7640268,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.723643687000731,
       "bytes": 1209843,
@@ -562,6 +556,12 @@
       "bytes": 7224601,
       "pages": 32,
       "shards": 3
+    },
+    {
+      "seconds": 1.2717098209977848,
+      "bytes": 11010766,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -179,3 +179,4 @@
 {"step": 212900, "loss": 0.6111213471740484, "lr": 0.0003, "grad_norm": 0.1558663111627959, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31161710459, "tokens_per_second": 148102.92774561647, "data_wait_s": 1.150800017057918e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 213000, "loss": 0.5666066184639931, "lr": 0.0003, "grad_norm": 0.1961237770430454, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31176111509, "tokens_per_second": 146116.2680647569, "data_wait_s": 1.881700154626742e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 213100, "loss": 0.6173598244786263, "lr": 0.0003, "grad_norm": 0.14808592038727653, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31190635549, "tokens_per_second": 147841.02489907615, "data_wait_s": 1.7596001271158457e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 213200, "loss": 0.5844620242714882, "lr": 0.0003, "grad_norm": 0.1403808011301817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 31205436021, "tokens_per_second": 129244.27535796976, "data_wait_s": 2.584400135674514e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f6c70f963f286fa4d8be08e8bbb732ea0d78cd473815a41380b1650070e25be9
-size 385872
+oid sha256:f7cf3712970ef12db198251d2ed33be2ced8969e6f6f7dfb9a411ce71183278c
+size 387990