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": 76918.49932590839,
-      "required_pages_per_second_upper_estimate": 0.0053214002588739045
+      "required_token_positions_per_second": 76736.93904024262,
+      "required_pages_per_second_upper_estimate": 0.0053088394970336435
     },
     {
       "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": 9614.812415738548,
-      "required_pages_per_second_upper_estimate": 0.0040255211783550215
+      "required_token_positions_per_second": 9592.117380030328,
+      "required_pages_per_second_upper_estimate": 0.004016019240830227
     },
     {
       "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": 19229.624831477096,
-      "required_pages_per_second_upper_estimate": 0.004005023733087577
+      "required_token_positions_per_second": 19184.234760060655,
+      "required_pages_per_second_upper_estimate": 0.003995570178228211
     },
     {
       "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": 28844.437247215646,
-      "required_pages_per_second_upper_estimate": 0.0017291799700650684
+      "required_token_positions_per_second": 28776.352140090985,
+      "required_pages_per_second_upper_estimate": 0.0017250983718528842
     },
     {
       "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": 28844.437247215646,
-      "required_pages_per_second_upper_estimate": 0.0022425281226999467
+      "required_token_positions_per_second": 28776.352140090985,
+      "required_pages_per_second_upper_estimate": 0.0022372348050957408
     },
     {
       "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": 28844.437247215646,
-      "required_pages_per_second_upper_estimate": 0.3699854702635375
+      "required_token_positions_per_second": 28776.352140090985,
+      "required_pages_per_second_upper_estimate": 0.36911214761343475
     }
   ],
   "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": 0.1143045802887741,
-    "producer_p95_seconds": 0.22746635200019227,
-    "gpu_update_seconds": 0.9585210404016068,
-    "producer_batches_per_second": 8.748555810044039,
-    "gpu_batches_per_second": 1.0432739166382974,
-    "producer_headroom": 8.385674817054934,
+    "producer_batch_seconds": 0.10123988209360846,
+    "producer_p95_seconds": 0.23909083399848896,
+    "gpu_update_seconds": 0.9607889097757122,
+    "producer_batches_per_second": 9.877530270880596,
+    "gpu_batches_per_second": 1.0408113476595409,
+    "producer_headroom": 9.490221540235963,
     "sustainable": true,
-    "buffer_seconds": 60.38682554530123,
+    "buffer_seconds": 60.529701315869865,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 415275.1241568944,
-    "upload_bytes_per_second": 4994423.116857663,
+    "generation_bytes_per_second": 455198.67840559874,
+    "upload_bytes_per_second": 4996323.83739647,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 89.10837453394055,
+    "estimated_full_shards_per_hour": 97.67503990293476,
     "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": 1.1105975430000399,
-      "bytes": 2661454,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.082202919998963,
       "bytes": 11730331,
@@ -562,6 +556,12 @@
       "bytes": 6314908,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0093370079994202,
+      "bytes": 2319025,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790824801708454555-41033672/train.jsonl

@@ -211,3 +211,4 @@
 {"step": 418200, "loss": 0.5855352282524109, "lr": 0.0003, "grad_norm": 0.1395451328877831, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61213494863, "tokens_per_second": 152492.80641446833, "data_wait_s": 1.1574000382097438e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 418300, "loss": 0.5724180936813354, "lr": 0.0003, "grad_norm": 0.15005547982988734, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61228169417, "tokens_per_second": 152253.36935442977, "data_wait_s": 1.5689001884311438e-05, "gpu_peak_gb": 89.37861251831055}
 {"step": 418400, "loss": 0.590329110622406, "lr": 0.0003, "grad_norm": 0.17341479540921204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61242727058, "tokens_per_second": 150676.7163422943, "data_wait_s": 1.2142998457420617e-05, "gpu_peak_gb": 89.37861251831055}
+{"step": 418500, "loss": 0.5561164021492004, "lr": 0.0003, "grad_norm": 0.15359413721278908, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 61257245218, "tokens_per_second": 150705.3470188801, "data_wait_s": 6.4209998527076095e-06, "gpu_peak_gb": 89.37861251831055}

runs/void-byte-v1/events.out.tfevents.1790824801.afc92292-ba09-4e51-8d82-bc1983323308-sdcdh.469.0.1790824801708454555-41033672

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9e1038ca65f35562cf62042442bf70f4920f5f0ebfd3a759735ac8475a66976f
-size 454431
+oid sha256:ace639a900eb3f43dc0d16120ffdf929e125eae7485147789c3fc26c1d966967
+size 456549