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": 77170.38413334845,
-      "required_pages_per_second_upper_estimate": 0.005338826234305883
+      "required_token_positions_per_second": 77391.67011016366,
+      "required_pages_per_second_upper_estimate": 0.005354135311636166
     },
     {
       "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": 9646.298016668556,
-      "required_pages_per_second_upper_estimate": 0.004038703541970302
+      "required_token_positions_per_second": 9673.958763770457,
+      "required_pages_per_second_upper_estimate": 0.004050284519159785
     },
     {
       "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": 19292.59603333711,
-      "required_pages_per_second_upper_estimate": 0.00401813897377275
+      "required_token_positions_per_second": 19347.917527540914,
+      "required_pages_per_second_upper_estimate": 0.004029660982089491
     },
     {
       "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": 28938.894050005663,
-      "required_pages_per_second_upper_estimate": 0.001734842511165143
+      "required_token_positions_per_second": 29021.87629131137,
+      "required_pages_per_second_upper_estimate": 0.0017398171698248745
     },
     {
       "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": 28938.894050005663,
-      "required_pages_per_second_upper_estimate": 0.0022498717236453035
+      "required_token_positions_per_second": 29021.87629131137,
+      "required_pages_per_second_upper_estimate": 0.002256323227906517
     },
     {
       "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": 28938.894050005663,
-      "required_pages_per_second_upper_estimate": 0.37119706070991476
+      "required_token_positions_per_second": 29021.87629131137,
+      "required_pages_per_second_upper_estimate": 0.3722614678019955
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.08067389404689607,
+    "producer_batch_seconds": 0.20843425083586453,
     "producer_p95_seconds": 0.21510461700017913,
-    "gpu_update_seconds": 0.9553924193587,
-    "producer_batches_per_second": 12.395583624842702,
-    "gpu_batches_per_second": 1.0466903229892095,
-    "producer_headroom": 11.842646628701553,
+    "gpu_update_seconds": 0.9526606661292025,
+    "producer_batches_per_second": 4.797675986503143,
+    "gpu_batches_per_second": 1.0496917061382873,
+    "producer_headroom": 4.5705572011741635,
     "sustainable": true,
-    "buffer_seconds": 60.1897224195981,
+    "buffer_seconds": 60.01762196613976,
     "buffer_drain_seconds": null,
-    "shard_target_mib": 32,
-    "upload_pages_per_commit": 173,
+    "shard_target_mib": 16,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 563370.1417194274,
-    "upload_bytes_per_second": 5036763.770853366,
+    "generation_bytes_per_second": 543485.9734106802,
+    "upload_bytes_per_second": 5125500.105311271,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 60.443058913646304,
+    "estimated_full_shards_per_hour": 116.61943818798355,
     "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": 1.8210042499995325,
-      "bytes": 7672821,
-      "pages": 32,
-      "shards": 4
-    },
     {
       "seconds": 1.2987783030002902,
       "bytes": 10941174,
@@ -562,6 +556,12 @@
       "bytes": 2424834,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 1.1364503580007295,
+      "bytes": 11698392,
+      "pages": 34,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -144,3 +144,4 @@
 {"step": 454600, "loss": 0.5783511996269226, "lr": 0.0003, "grad_norm": 0.14675936160243108, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66539606627, "tokens_per_second": 151291.57431869392, "data_wait_s": 1.3990998922963627e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 454700, "loss": 0.5707882046699524, "lr": 0.0003, "grad_norm": 0.1699751712767245, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66554171824, "tokens_per_second": 151386.46710184007, "data_wait_s": 1.4927998563507572e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 454800, "loss": 0.5946531891822815, "lr": 0.0003, "grad_norm": 0.15826047422264783, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66568790965, "tokens_per_second": 152278.17276115512, "data_wait_s": 1.2017000699415803e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 454900, "loss": 0.5882903337478638, "lr": 0.0003, "grad_norm": 0.18120452433200254, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 66583435758, "tokens_per_second": 152681.19558098027, "data_wait_s": 9.585999578121118e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c487768688b2864bb95ae69615d5354fe723a4132ed8493750a1dc7601dec339
-size 310959
+oid sha256:f29e75c5d671438363179e40ab10418485cfd9d77aeadfd0f7b0467f68110cfd
+size 313077