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": 75146.88202418211,
-      "required_pages_per_second_upper_estimate": 0.005198835663221994
+      "required_token_positions_per_second": 75441.4904095358,
+      "required_pages_per_second_upper_estimate": 0.005219217354906397
     },
     {
       "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": 9393.360253022764,
-      "required_pages_per_second_upper_estimate": 0.003932803782272944
+      "required_token_positions_per_second": 9430.186301191976,
+      "required_pages_per_second_upper_estimate": 0.00394822207962606
     },
     {
       "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": 18786.720506045527,
-      "required_pages_per_second_upper_estimate": 0.0039127784422727
+      "required_token_positions_per_second": 18860.37260238395,
+      "required_pages_per_second_upper_estimate": 0.003928118231603597
     },
     {
       "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": 28180.08075906829,
-      "required_pages_per_second_upper_estimate": 0.0016893528130142554
+      "required_token_positions_per_second": 28290.55890357593,
+      "required_pages_per_second_upper_estimate": 0.0016959758090871293
     },
     {
       "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": 28180.08075906829,
-      "required_pages_per_second_upper_estimate": 0.0021908773279418537
+      "required_token_positions_per_second": 28290.55890357593,
+      "required_pages_per_second_upper_estimate": 0.002199466517735321
     },
     {
       "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": 28180.08075906829,
-      "required_pages_per_second_upper_estimate": 0.36146381856400367
+      "required_token_positions_per_second": 28290.55890357593,
+      "required_pages_per_second_upper_estimate": 0.3628809135795581
     }
   ],
   "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": 0.5958665545621784,
-    "producer_p95_seconds": 0.26070974499816657,
-    "gpu_update_seconds": 0.9811185509503174,
-    "producer_batches_per_second": 1.6782281071888059,
-    "gpu_batches_per_second": 1.0192448191213936,
-    "producer_headroom": 1.6465407286891751,
+    "producer_batch_seconds": 0.06999284335952893,
+    "producer_p95_seconds": 0.24402895200182684,
+    "gpu_update_seconds": 0.9772871612128274,
+    "producer_batches_per_second": 14.287174973923365,
+    "gpu_batches_per_second": 1.0232407010841988,
+    "producer_headroom": 13.962672672016518,
     "sustainable": true,
-    "buffer_seconds": 60.82935015891968,
+    "buffer_seconds": 60.5918039951953,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 365415.8920625823,
-    "upload_bytes_per_second": 5304161.810963294,
+    "generation_bytes_per_second": 365410.11048667855,
+    "upload_bytes_per_second": 5356788.565272251,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 78.40974399002172,
+    "estimated_full_shards_per_hour": 78.40850339842098,
     "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": 1.3613106869997864,
-      "bytes": 11006649,
-      "pages": 33,
-      "shards": 4
-    },
     {
       "seconds": 0.8713928039978782,
       "bytes": 1169048,
@@ -562,6 +556,12 @@
       "bytes": 2380354,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.6319502810001723,
+      "bytes": 16693932,
+      "pages": 40,
+      "shards": 4
     }
   ],
   "cache_key": {

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

@@ -245,3 +245,4 @@
 {"step": 744500, "loss": 0.5162256360054016, "lr": 5e-05, "grad_norm": 0.24195962382761704, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108974336951, "tokens_per_second": 151422.97107831604, "data_wait_s": 1.6177997167687863e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 744600, "loss": 0.498966246843338, "lr": 5e-05, "grad_norm": 0.22884791616754208, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 108989067899, "tokens_per_second": 152174.10579863703, "data_wait_s": 1.4053999620955437e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 744700, "loss": 0.5444483757019043, "lr": 5e-05, "grad_norm": 0.24081576552242007, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109003738729, "tokens_per_second": 150383.42225768467, "data_wait_s": 1.0996998753398657e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 744800, "loss": 0.5238848328590393, "lr": 5e-05, "grad_norm": 0.23703562577114629, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 109018512323, "tokens_per_second": 152851.77457409486, "data_wait_s": 1.4173001545714214e-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 sha256:5109e573656f901f25a33907623b2d539e3ec87ef9bcf869f55f66964edd1b23
-size 526443
+oid sha256:7dbc7887ef3885fe1b0be9cf113c713f1d8047f342322a30225486a52da1de43
+size 528561