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": 76549.46681962843,
-      "required_pages_per_second_upper_estimate": 0.005295869733816071
+      "required_token_positions_per_second": 77654.17740842642,
+      "required_pages_per_second_upper_estimate": 0.005372296175630833
     },
     {
       "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": 9568.683352453554,
-      "required_pages_per_second_upper_estimate": 0.004006207902841974
+      "required_token_positions_per_second": 9706.772176053302,
+      "required_pages_per_second_upper_estimate": 0.0040640228096606446
     },
     {
       "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": 19137.36670490711,
-      "required_pages_per_second_upper_estimate": 0.003985808798333459
+      "required_token_positions_per_second": 19413.544352106604,
+      "required_pages_per_second_upper_estimate": 0.004043329318950776
     },
     {
       "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": 28706.050057360662,
-      "required_pages_per_second_upper_estimate": 0.0017208838693382662
+      "required_token_positions_per_second": 29120.31652815991,
+      "required_pages_per_second_upper_estimate": 0.001745718511714406
     },
     {
       "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": 28706.050057360662,
-      "required_pages_per_second_upper_estimate": 0.0022317691273896403
+      "required_token_positions_per_second": 29120.31652815991,
+      "required_pages_per_second_upper_estimate": 0.002263976523328649
     },
     {
       "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": 28706.050057360662,
-      "required_pages_per_second_upper_estimate": 0.36821038798066547
+      "required_token_positions_per_second": 29120.31652815991,
+      "required_pages_per_second_upper_estimate": 0.37352415346339723
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05513125057035495,
-    "producer_p95_seconds": 0.1722166190002099,
-    "gpu_update_seconds": 0.9631419141523665,
-    "producer_batches_per_second": 18.138532858489476,
-    "gpu_batches_per_second": 1.0382685929311584,
-    "producer_headroom": 17.46998125724115,
+    "producer_batch_seconds": 0.09812022645315466,
+    "producer_p95_seconds": 0.20461092699952133,
+    "gpu_update_seconds": 0.9494402292387121,
+    "producer_batches_per_second": 10.191578598500564,
+    "gpu_batches_per_second": 1.053252189241895,
+    "producer_headroom": 9.676294720864728,
     "sustainable": true,
-    "buffer_seconds": 60.67794059159909,
+    "buffer_seconds": 60.76417467127757,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 396544.71354545205,
-    "upload_bytes_per_second": 2625602.23827878,
+    "generation_bytes_per_second": 427555.3797785834,
+    "upload_bytes_per_second": 2613961.9997697156,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.08926443836852,
+    "estimated_full_shards_per_hour": 91.74343152063491,
     "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": [
@@ -358,6 +358,12 @@
       "bytes": 5842307,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.3798525270003665,
+      "bytes": 2740249,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -69,3 +69,4 @@
 {"step": 651900, "loss": 0.5858619809150696, "lr": 0.0001, "grad_norm": 0.17948010360868635, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95420643561, "tokens_per_second": 153603.0674927566, "data_wait_s": 9.302000762545504e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 652000, "loss": 0.5466394424438477, "lr": 0.0001, "grad_norm": 0.19738119206916308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95435357325, "tokens_per_second": 153580.7515054078, "data_wait_s": 1.070100006472785e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 652100, "loss": 0.5375281572341919, "lr": 0.0001, "grad_norm": 0.1803161872897369, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95449862098, "tokens_per_second": 151957.2287803522, "data_wait_s": 1.1095999980170745e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 652200, "loss": 0.5419741272926331, "lr": 0.0001, "grad_norm": 0.1745661499664961, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 95464388264, "tokens_per_second": 151707.58956482777, "data_wait_s": 2.0442999812075868e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e5dde38f05f49f1a4da8120e6a93ce4548978b0d848d4dd0f8ce04a7c76f28f9
-size 151326
+oid sha256:b8e5187fb0ac7303725e56cc7b624cd641bb8f4bb1958f6d5574b71d4223fe74
+size 153444