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": 76128.58543616634,
-      "required_pages_per_second_upper_estimate": 0.005266752183128808
+      "required_token_positions_per_second": 76059.38067779403,
+      "required_pages_per_second_upper_estimate": 0.005261964437367449
     },
     {
       "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": 9516.073179520792,
-      "required_pages_per_second_upper_estimate": 0.003984181122060367
+      "required_token_positions_per_second": 9507.422584724254,
+      "required_pages_per_second_upper_estimate": 0.003980559298664018
     },
     {
       "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": 19032.146359041584,
-      "required_pages_per_second_upper_estimate": 0.00396389417513679
+      "required_token_positions_per_second": 19014.84516944851,
+      "required_pages_per_second_upper_estimate": 0.003960290793607605
     },
     {
       "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": 28548.219538562378,
-      "required_pages_per_second_upper_estimate": 0.0017114221707295572
+      "required_token_positions_per_second": 28522.267754172764,
+      "required_pages_per_second_upper_estimate": 0.0017098664008814801
     },
     {
       "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": 28548.219538562378,
-      "required_pages_per_second_upper_estimate": 0.0022194984987761583
+      "required_token_positions_per_second": 28522.267754172764,
+      "required_pages_per_second_upper_estimate": 0.0022174808616897013
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28548.219538562378,
-      "required_pages_per_second_upper_estimate": 0.36618590755072894
+      "required_token_positions_per_second": 28522.267754172764,
+      "required_pages_per_second_upper_estimate": 0.36585302592543406
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.12767598046869466,
-    "producer_p95_seconds": 0.28170336699986365,
-    "gpu_update_seconds": 0.968466701142382,
-    "producer_batches_per_second": 7.8323267722639,
-    "gpu_batches_per_second": 1.0325600238195305,
-    "producer_headroom": 7.5853476714035795,
+    "producer_batch_seconds": 0.07634836460158567,
+    "producer_p95_seconds": 0.2554694369973731,
+    "gpu_update_seconds": 0.969347887702763,
+    "producer_batches_per_second": 13.09785750118387,
+    "gpu_batches_per_second": 1.0316213742105311,
+    "producer_headroom": 12.696380502204375,
     "sustainable": true,
-    "buffer_seconds": 60.04493547082768,
+    "buffer_seconds": 60.09956903757131,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 279314.8182255779,
-    "upload_bytes_per_second": 5315256.382291279,
+    "generation_bytes_per_second": 279584.34545108426,
+    "upload_bytes_per_second": 5278606.065685195,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 59.93445787501814,
+    "estimated_full_shards_per_hour": 59.992292143339114,
     "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": 1.1391658569991705,
-      "bytes": 11247930,
-      "pages": 27,
-      "shards": 3
-    },
     {
       "seconds": 1.2541976089996751,
       "bytes": 7910419,
@@ -562,6 +556,12 @@
       "bytes": 1047147,
       "pages": 31,
       "shards": 1
+    },
+    {
+      "seconds": 1.6690692930023943,
+      "bytes": 11069831,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -280,3 +280,4 @@
 {"step": 748000, "loss": 0.517157256603241, "lr": 5e-05, "grad_norm": 0.2323156664259303, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109486659350, "tokens_per_second": 150072.0760592924, "data_wait_s": 9.571002010488883e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 748100, "loss": 0.47873765230178833, "lr": 5e-05, "grad_norm": 0.23634672132481224, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109501232079, "tokens_per_second": 149199.96091553674, "data_wait_s": 1.2941000022692606e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 748200, "loss": 0.5426670908927917, "lr": 5e-05, "grad_norm": 0.2458641982590552, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109515797465, "tokens_per_second": 149982.21513129355, "data_wait_s": 1.3760000001639128e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 748300, "loss": 0.5866421461105347, "lr": 5e-05, "grad_norm": 0.2558668659257247, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109530450067, "tokens_per_second": 151991.75967971134, "data_wait_s": 1.6761001461418346e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ec8efc4a2390349eb00c1a7324367669eb91e0cef4e42dcb3cf3aa0072e9f2b3
-size 601356
+oid sha256:03bb4166b83d2e516fd8c4b32a30dec217c7d654caf2327205480a52c8f2048f
+size 603474