Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 74143.30127656019,
-      "required_pages_per_second_upper_estimate": 0.0064265534401731494
+      "required_token_positions_per_second": 74378.58776369953,
+      "required_pages_per_second_upper_estimate": 0.006446947476550237
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9267.912659570024,
-      "required_pages_per_second_upper_estimate": 0.003697108927545087
+      "required_token_positions_per_second": 9297.323470462441,
+      "required_pages_per_second_upper_estimate": 0.0037088413397408813
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18535.825319140047,
-      "required_pages_per_second_upper_estimate": 0.004287936968199208
+      "required_token_positions_per_second": 18594.646940924882,
+      "required_pages_per_second_upper_estimate": 0.004301544315173944
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27803.73797871007,
-      "required_pages_per_second_upper_estimate": 0.0016931017679640782
+      "required_token_positions_per_second": 27891.970411387323,
+      "required_pages_per_second_upper_estimate": 0.0016984746601943252
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 27803.73797871007,
-      "required_pages_per_second_upper_estimate": 0.0021404087152738483
+      "required_token_positions_per_second": 27891.970411387323,
+      "required_pages_per_second_upper_estimate": 0.002147201092184349
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 21,
-      "required_token_positions_per_second": 27803.73797871007,
-      "required_pages_per_second_upper_estimate": 0.3452293725704964
+      "required_token_positions_per_second": 27891.970411387323,
+      "required_pages_per_second_upper_estimate": 0.3463249240893915
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.40416656830490183,
-    "producer_p95_seconds": 0.4593587560011656,
-    "gpu_update_seconds": 0.9943986675881739,
-    "producer_batches_per_second": 2.4742274062747405,
-    "gpu_batches_per_second": 1.0056328840679278,
-    "producer_headroom": 2.460368436109745,
+    "producer_batch_seconds": 0.14202504810936034,
+    "producer_p95_seconds": 0.44807584999944083,
+    "gpu_update_seconds": 0.9912530234404767,
+    "producer_batches_per_second": 7.04101152093955,
+    "gpu_batches_per_second": 1.0088241612914974,
+    "producer_headroom": 6.979423958210559,
     "sustainable": true,
-    "buffer_seconds": 60.658318722878604,
+    "buffer_seconds": 60.46643442986908,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 311420.5175339137,
-    "upload_bytes_per_second": 5499434.006551638,
+    "generation_bytes_per_second": 311930.23672893003,
+    "upload_bytes_per_second": 5398958.192868696,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.82359356415805,
+    "estimated_full_shards_per_hour": 66.93296743775296,
     "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": 1.0217303430035827,
-      "bytes": 7986291,
-      "pages": 39,
-      "shards": 3
-    },
     {
       "seconds": 0.8759991519982577,
       "bytes": 805335,
@@ -562,6 +556,12 @@
       "bytes": 10136241,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1429420250060502,
+      "bytes": 943208,
+      "pages": 27,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -404,3 +404,4 @@
 {"step": 355400, "loss": 0.6366513706743717, "lr": 0.0003, "grad_norm": 0.18563034325243508, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52020506515, "tokens_per_second": 147607.4892225609, "data_wait_s": 9.976000001188368e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 355500, "loss": 0.5926955752074718, "lr": 0.0003, "grad_norm": 0.14963007934827288, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52035227901, "tokens_per_second": 140887.84228410333, "data_wait_s": 1.6331003280356526e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 355600, "loss": 0.5659326519817114, "lr": 0.0003, "grad_norm": 0.1694171388252178, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52049949793, "tokens_per_second": 149057.5964689041, "data_wait_s": 1.107400021282956e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 355700, "loss": 0.580079585313797, "lr": 0.0003, "grad_norm": 0.14113312019679025, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 52064543552, "tokens_per_second": 148225.1154835935, "data_wait_s": 9.800001862458885e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6cb5b966af88533db8513ce1155dd1ce1acf43c47891d08597112fe5f98d4a67
-size 866337
+oid sha256:20b85de7842ca73e1423b9d34ce5ecb224ea69f562d6064393165904a576d0c6
+size 868455