diff --git a/test/sample_000000/board.png b/test/sample_000000/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..ae9deba5c3d7a1ca22600a7301b3a6577c7360b6
--- /dev/null
+++ b/test/sample_000000/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:04d0d38969c3641b4f8e7aeb94ea30243c214e82a98a6068d852ec08188139aa
+size 38284
diff --git a/test/sample_000000/metadata.json b/test/sample_000000/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..a3d2b6b2573157d5c65c7bbf6b04dbba2a735f35
--- /dev/null
+++ b/test/sample_000000/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000000",
+  "puzzle_id": "TcdsK",
+  "task": "task2",
+  "fen": "4Q3/p5pp/2p1p1k1/2Pp4/3n2B1/8/P2q2rP/5R1K b - - 5 28",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Kg5",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/test/sample_000001/board.png b/test/sample_000001/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..e456a512ba7e9270099da3bfae3f55b4660494bb
--- /dev/null
+++ b/test/sample_000001/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d986e9109a01369952a18fa306c8baeac10c9fd76f8ee280a7f02d913bfd44c
+size 49483
diff --git a/test/sample_000001/metadata.json b/test/sample_000001/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..45a05e16bb769861e034cc33dcc2ee8d555c9ad2
--- /dev/null
+++ b/test/sample_000001/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000001",
+  "puzzle_id": "OARnq",
+  "task": "task2",
+  "fen": "3q1rk1/1p1bbppp/p3p3/1n1pP3/3N1P2/3QB2P/PPB3P1/2R3K1 b - - 4 22",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Nxd4",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/test/sample_000002/board.png b/test/sample_000002/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..539013e8c935dfff7f2eb368a3d159eba50c130b
--- /dev/null
+++ b/test/sample_000002/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b3708885ac0ddd2adf2ed87c313ca02ac88dbfc94e8689696f3723faf0bf1e8
+size 34412
diff --git a/test/sample_000002/metadata.json b/test/sample_000002/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..dcd5815c5399754afb15cc1a5be8d8dc64a9ba9b
--- /dev/null
+++ b/test/sample_000002/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000002",
+  "puzzle_id": "80JZP",
+  "task": "task2",
+  "fen": "8/2Kbk3/1B1p4/2pPp3/2B1Pp2/pP6/Pr3R2/8 w - - 1 51",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Rxb2",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/test/sample_000003/board.png b/test/sample_000003/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..85c533cb3f8f65d70fe4f633baedb6eb053a1a16
--- /dev/null
+++ b/test/sample_000003/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c64c34574475026506dcee47cf2cc2ad1483886e4830b81ba9f2bbc4b536957
+size 38953
diff --git a/test/sample_000003/metadata.json b/test/sample_000003/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..90df35d0940c364fee9d538e9c2d1dce5b8b8bbf
--- /dev/null
+++ b/test/sample_000003/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000003",
+  "puzzle_id": "Ol6gP",
+  "task": "task2",
+  "fen": "6k1/4bppp/2b1p3/1pNpP3/3P4/P3B3/r4PPP/2R3K1 w - - 4 27",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Nxe6",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000000/board.png b/train/sample_000000/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..dd07deb8eca4646af79ee758c4ae25a0dfec8495
--- /dev/null
+++ b/train/sample_000000/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7627083959d5e985257fa097021a3387f47cb1d3bac6a6a666b65015384a8658
+size 42045
diff --git a/train/sample_000000/metadata.json b/train/sample_000000/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..332c6dab93490f39894a498a69db57c8ec4b0bad
--- /dev/null
+++ b/train/sample_000000/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000000",
+  "puzzle_id": "2GDeK",
+  "task": "task2",
+  "fen": "r1b2rk1/ppRq3p/3p2p1/3PPp2/8/3B1NP1/2Q2K1P/8 b - - 1 30",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Qe8",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000001/board.png b/train/sample_000001/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..8fbd28d28813dd39cc2ddbf98383cc1817a76356
--- /dev/null
+++ b/train/sample_000001/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f902eeea6209bca731ca5f352adb5a4ef4c488011f50f03d9174eb149027b9e
+size 51548
diff --git a/train/sample_000001/metadata.json b/train/sample_000001/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..9c3db9a07f5cb576a7b5e02aea7f0df42cad03a1
--- /dev/null
+++ b/train/sample_000001/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000001",
+  "puzzle_id": "0wbi8",
+  "task": "task2",
+  "fen": "2rqk2r/pp1nppbp/5np1/3N2B1/3P4/4PQ1P/PP3PP1/R3KB1R w KQk - 1 11",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Bb5",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000002/board.png b/train/sample_000002/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..f618d3cc5158c7612a4340849bf2e56a3f678c6b
--- /dev/null
+++ b/train/sample_000002/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:155b9cc92f94024a58ff76662df8db761deeb8f16d5ec5dc32982c04d9bbd64b
+size 42182
diff --git a/train/sample_000002/metadata.json b/train/sample_000002/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..cdb8235ca62b5a016942820cbebf917d50c6c5c2
--- /dev/null
+++ b/train/sample_000002/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000002",
+  "puzzle_id": "D2hb6",
+  "task": "task2",
+  "fen": "3r1qrk/5p1p/p3pPp1/1p6/2p4Q/3B3R/PPP2PPP/6K1 b - - 1 26",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "h5",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000003/board.png b/train/sample_000003/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..3b29cd68880bcd4916a5c643871cf4a267d27463
--- /dev/null
+++ b/train/sample_000003/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3577eeb8b77c567f9077f27b44843ed33c0adff9ec25e80572b8a5f4162a8f4d
+size 24560
diff --git a/train/sample_000003/metadata.json b/train/sample_000003/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..9af6ffdb3ec95c0e995a196698ce77f31f580a67
--- /dev/null
+++ b/train/sample_000003/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000003",
+  "puzzle_id": "pE4Sx",
+  "task": "task2",
+  "fen": "8/5p2/1r6/2K1k3/1P1N4/7P/8/8 b - - 2 49",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Rb8",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000004/board.png b/train/sample_000004/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..61dc5df9619e33ffb623240292c386339e32fd73
--- /dev/null
+++ b/train/sample_000004/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b54ae6402a67b36550c6922162697f2282e4323a6b9bce98782080e65d031e7a
+size 43986
diff --git a/train/sample_000004/metadata.json b/train/sample_000004/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..19c18ec51405228846812b42095c50c84bc03ce4
--- /dev/null
+++ b/train/sample_000004/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000004",
+  "puzzle_id": "JkqQg",
+  "task": "task2",
+  "fen": "6k1/p2r1p2/3Np2p/2P1b3/5P2/1P3Rpb/P2QP1Bq/2R2K2 w - - 4 31",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Bxh3",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000005/board.png b/train/sample_000005/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..a3a500162ae7ba5f7884bb2aaa0098559c0b2637
--- /dev/null
+++ b/train/sample_000005/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51b9c8ec74cdb96e3f5101fc0bf6f0b0d427ddce92f6ec2cc858ee6cb50f5307
+size 50265
diff --git a/train/sample_000005/metadata.json b/train/sample_000005/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..44c72d31c3d6aa5811cd26d66b81dbab0e4a37eb
--- /dev/null
+++ b/train/sample_000005/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000005",
+  "puzzle_id": "ycKtk",
+  "task": "task2",
+  "fen": "3rr2k/pQp1q1bp/5pp1/5b2/8/2Pn1NP1/PP2PPBP/R1B1R1K1 w - - 3 19",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "exd3",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000006/board.png b/train/sample_000006/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..5254ae02ae19418dc9c89fe05b1766f2d959ac7e
--- /dev/null
+++ b/train/sample_000006/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3caa4056f3c0b43001911ac4a9b029ace1a7db0decf83db6bfdca44424f450b3
+size 43945
diff --git a/train/sample_000006/metadata.json b/train/sample_000006/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..02e209a4f512c08792da0d56f0f722227d2ec063
--- /dev/null
+++ b/train/sample_000006/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000006",
+  "puzzle_id": "jAI6L",
+  "task": "task2",
+  "fen": "1qk4r/1p3ppr/p1p3p1/3p4/Q5n1/P3P2P/1P1B1PP1/R2R2K1 b - - 11 23",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Rxh3",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000007/board.png b/train/sample_000007/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..58a257a8034fe994edcebe0b4ce772d80c880856
--- /dev/null
+++ b/train/sample_000007/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:62281e60ebb30301a0c36bfefe8036b97ebf069f0fafeb31a554a67bf46fdac6
+size 39852
diff --git a/train/sample_000007/metadata.json b/train/sample_000007/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..166f9fc55f553060451214750421ec411fc39d85
--- /dev/null
+++ b/train/sample_000007/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000007",
+  "puzzle_id": "YwFEa",
+  "task": "task2",
+  "fen": "3r3k/p2q2p1/1p3p1R/2p5/2P1bP2/1P1r4/PB5R/2K1Q3 b - - 0 34",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Kg8",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000008/board.png b/train/sample_000008/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..34b67381bfb9f4743dbcf11627718a655fcd5587
--- /dev/null
+++ b/train/sample_000008/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a880bba0545483636e27ef0c6b5f95a6c1ca353d73d9bd63f57d8e02cbfcffc
+size 39867
diff --git a/train/sample_000008/metadata.json b/train/sample_000008/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..1d951d07c69557aad049dc44762b111e634ce975
--- /dev/null
+++ b/train/sample_000008/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000008",
+  "puzzle_id": "PX1st",
+  "task": "task2",
+  "fen": "2Q5/5rk1/5n1p/p5p1/PpN1P3/1P6/4Bq1P/6RK w - - 2 39",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Rg2",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000009/board.png b/train/sample_000009/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..61e126454c5692572a34a61a93f2f7c0b86d4c1c
--- /dev/null
+++ b/train/sample_000009/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b560945df8cd6f3e6ba357a1b5d80e8810448767f1dd4e234ecdd5b974f08609
+size 33006
diff --git a/train/sample_000009/metadata.json b/train/sample_000009/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..aa8db6eff2d790702457d4621a01dca6e372a5fa
--- /dev/null
+++ b/train/sample_000009/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000009",
+  "puzzle_id": "wxfOn",
+  "task": "task2",
+  "fen": "3r1nk1/pp2R1p1/2pn4/6p1/8/1P6/P4P1P/4R1K1 b - - 1 32",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Rd7",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000010/board.png b/train/sample_000010/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..884b50ff53570863192fe9605ff6d6f57c67db44
--- /dev/null
+++ b/train/sample_000010/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7eb33c0f2f956acadb9ae2a48db8664647dd7a9acb7d1054432688fb968f8a2e
+size 47224
diff --git a/train/sample_000010/metadata.json b/train/sample_000010/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..37c5f6a3ffc729347e8a94ab449cb8b54d9929e8
--- /dev/null
+++ b/train/sample_000010/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000010",
+  "puzzle_id": "jsdqV",
+  "task": "task2",
+  "fen": "2n4k/2p2b1p/p2qNbp1/1p1P1p2/1P6/PBP1Q2P/5PK1/2B5 w - - 8 28",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "c4",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000011/board.png b/train/sample_000011/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..46659990841755a888143277310313d09da0b360
--- /dev/null
+++ b/train/sample_000011/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a26b438e87c228a3ccccd999c94b9b3838c845599fe50c1116c68e262aee8f92
+size 38847
diff --git a/train/sample_000011/metadata.json b/train/sample_000011/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..d220283d6fa54a6bd88c89533262df1aa9ba1eea
--- /dev/null
+++ b/train/sample_000011/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000011",
+  "puzzle_id": "n3Uo4",
+  "task": "task2",
+  "fen": "r3r1k1/1p3ppp/8/p2p4/P7/2P1PPq1/B2Q4/3RK2R w K - 0 25",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Qf2",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000012/board.png b/train/sample_000012/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..76e275c427e7bceebb0449ca9d28a21239107121
--- /dev/null
+++ b/train/sample_000012/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3b3358209dc8b81adbe0966c8618713dbb86424c303ecc3a33443e64fd27015d
+size 28620
diff --git a/train/sample_000012/metadata.json b/train/sample_000012/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..6a0a2ada81734803ef586cb824395df7eb11acfd
--- /dev/null
+++ b/train/sample_000012/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000012",
+  "puzzle_id": "g7LS8",
+  "task": "task2",
+  "fen": "6B1/6p1/7p/5P2/1p2nkPK/8/P7/8 w - - 3 50",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Kh5",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000013/board.png b/train/sample_000013/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..0032ab80b5f9571eb5b6bdbc45d42a2165c2bc0c
--- /dev/null
+++ b/train/sample_000013/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ae90702d99c13946824d943c4b45e6453dbff6e936eb437820b2571129920f2
+size 27120
diff --git a/train/sample_000013/metadata.json b/train/sample_000013/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..fb2683a13b89e75eff89597567af58d4ecace60c
--- /dev/null
+++ b/train/sample_000013/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000013",
+  "puzzle_id": "M04Bm",
+  "task": "task2",
+  "fen": "8/8/5ppp/pk6/2p2PPP/2P5/1K6/8 b - - 1 46",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Ka4",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000014/board.png b/train/sample_000014/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..d833ba727257e5b8a212095896a4d01d14d638e6
--- /dev/null
+++ b/train/sample_000014/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0af877bec4c229dc726db811f1b5725df27a9aa1303781a212878f01c7366110
+size 46274
diff --git a/train/sample_000014/metadata.json b/train/sample_000014/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..d7cbcbf2bbbfd459271c20a6bf7fe6c8dc6c41b8
--- /dev/null
+++ b/train/sample_000014/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000014",
+  "puzzle_id": "xxix0",
+  "task": "task2",
+  "fen": "4rrk1/pp4pp/3q2n1/3pN2n/3P1P2/2PQ2R1/PP1N3P/5R1K w - - 7 26",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Rh3",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000015/board.png b/train/sample_000015/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..b77f27aef9fe079a05d9f1986c4fe0ed8bf61639
--- /dev/null
+++ b/train/sample_000015/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e6da8aea4abc946aac90fb748bc027ee05065e0188b16bb6f0d173336e7eb99
+size 42892
diff --git a/train/sample_000015/metadata.json b/train/sample_000015/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..d14d81f6583bee5e43f077c243415013581e3621
--- /dev/null
+++ b/train/sample_000015/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000015",
+  "puzzle_id": "QfMBq",
+  "task": "task2",
+  "fen": "3q4/5rpk/4R1np/bp3p1Q/pNr5/P6P/1P3PP1/4R1K1 b - - 13 33",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Nf4",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000016/board.png b/train/sample_000016/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..f13f55a6aca183e7074e20a7620cb09cde5d80a5
--- /dev/null
+++ b/train/sample_000016/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:83c1660f7438612f1b8fd76fcec77a33fa332268c2315123afb2a27d46c9afaf
+size 30161
diff --git a/train/sample_000016/metadata.json b/train/sample_000016/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..7bbff49a9c89f0a9cdfeef4dde5417eb93010cd8
--- /dev/null
+++ b/train/sample_000016/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000016",
+  "puzzle_id": "ddNXk",
+  "task": "task2",
+  "fen": "8/2p5/8/p1k2p1p/2P5/3K1P1P/P1N2b2/8 w - - 0 37",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Ne3",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000017/board.png b/train/sample_000017/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..593a410ed0158cd3484dbfdccc3d756573106ee0
--- /dev/null
+++ b/train/sample_000017/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47e4036d2ee3e93df14227b0f3a3be179eedd759daad55fea1d5016e61c4ce44
+size 40200
diff --git a/train/sample_000017/metadata.json b/train/sample_000017/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..01d8a8df271d91e36c567167927c14f3e434cad6
--- /dev/null
+++ b/train/sample_000017/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000017",
+  "puzzle_id": "lP41O",
+  "task": "task2",
+  "fen": "rn4k1/p5pp/1P2p3/3pP3/q7/P1QPP3/6PP/5RK1 b - - 0 22",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "axb6",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000018/board.png b/train/sample_000018/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..dc141b2193d3e1130aa79f30b4bfdb22eafaa48d
--- /dev/null
+++ b/train/sample_000018/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:78ac6a9235cdd6310b0ee169bb45674fdc58722a64755659e6398c0446ae1ae4
+size 28453
diff --git a/train/sample_000018/metadata.json b/train/sample_000018/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..6d0e9c1bbadef112402f88bda5da20a779806b11
--- /dev/null
+++ b/train/sample_000018/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000018",
+  "puzzle_id": "Q4Gpp",
+  "task": "task2",
+  "fen": "2Rr1k2/2R2pp1/4p2p/8/P7/4K1P1/3r3P/8 b - - 0 54",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Rxc8",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000019/board.png b/train/sample_000019/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..f2f113bc7daeb8d2a0b8bb4bfcb2fd91393ba169
--- /dev/null
+++ b/train/sample_000019/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a74e16ddbfeb397379d808a5ef7d25d21b73e7d4256c51c6c864198e166499d
+size 52052
diff --git a/train/sample_000019/metadata.json b/train/sample_000019/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..a68f000aa90b723a4d9b8aa3ee8cb141191a14eb
--- /dev/null
+++ b/train/sample_000019/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000019",
+  "puzzle_id": "20jMs",
+  "task": "task2",
+  "fen": "r4rk1/1bq1pp1p/p5p1/npp5/6n1/1NP1PN2/PPB1QPPP/3R1RK1 w - - 6 19",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "h3",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000020/board.png b/train/sample_000020/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..c42d31ab20b9e34f502389200600ba00c5572b17
--- /dev/null
+++ b/train/sample_000020/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8e6588ab4c3b7f8bea52be503dcbdd2a0bb18232d471faf3cd68ee6798a30dc
+size 45169
diff --git a/train/sample_000020/metadata.json b/train/sample_000020/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..7ee7a5e7a438fd078fbf817ce36c1e9758e42a0f
--- /dev/null
+++ b/train/sample_000020/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000020",
+  "puzzle_id": "6Ixyo",
+  "task": "task2",
+  "fen": "rb4r1/1p3p1k/3p3B/qN2pP1N/2P1P2P/4Q3/6BK/4n3 b - - 1 35",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Nxg2",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000021/board.png b/train/sample_000021/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..047c7195b14d963f1a5a2ae58cb10c35b4dd2e73
--- /dev/null
+++ b/train/sample_000021/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e02b7821ffe2475b0c51694b828c6fa0b3de110ae9a7a451a7fa102347f37cbd
+size 50884
diff --git a/train/sample_000021/metadata.json b/train/sample_000021/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..5dd57f6c01218821fc36413a1ac784b63452a5dd
--- /dev/null
+++ b/train/sample_000021/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000021",
+  "puzzle_id": "UQObr",
+  "task": "task2",
+  "fen": "r7/1b1n1pbk/3Np1p1/q1P3Pp/4PPnN/1P5P/1PQ5/1K1R2BR w - - 1 25",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "hxg4",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000022/board.png b/train/sample_000022/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..e40fcac9d9ed85b51200b6a584cca81a6a969b7f
--- /dev/null
+++ b/train/sample_000022/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f68e806e3fbe631d6dc880cbea5ebbf8a1a2757b4f81ca5e75bc8302f281df4
+size 39427
diff --git a/train/sample_000022/metadata.json b/train/sample_000022/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..e2ff645f42e3e6a9f5a730ba10702bc438a3072a
--- /dev/null
+++ b/train/sample_000022/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000022",
+  "puzzle_id": "2dgPx",
+  "task": "task2",
+  "fen": "6k1/2p4p/3p1r2/2PPpp2/4b3/2Q1P1Pq/1B3R1P/6K1 b - - 0 27",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "dxc5",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000023/board.png b/train/sample_000023/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..d5dcbf488c14f52c649a5fc7ca510ab02c702f86
--- /dev/null
+++ b/train/sample_000023/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54b3aa545df08644d80857035313a2beb6f637466cad0b964915b7a2cb9cb66d
+size 36250
diff --git a/train/sample_000023/metadata.json b/train/sample_000023/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..3c40bbae1e18baabd631b8a16fe17e431c7eb158
--- /dev/null
+++ b/train/sample_000023/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000023",
+  "puzzle_id": "wGije",
+  "task": "task2",
+  "fen": "8/3b1p2/5Nkp/3PK1pN/3QP3/5q2/7P/8 w - - 4 44",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Nf4+",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000024/board.png b/train/sample_000024/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..43899042929e7ed6684c5c9e6caea03a196b3dba
--- /dev/null
+++ b/train/sample_000024/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d65c4db1dbdca1dba882f44a86d9f6d5db4f8553aa32dff46f9f80ce069a1167
+size 29612
diff --git a/train/sample_000024/metadata.json b/train/sample_000024/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..cf7c0619dc2838649b24774a914fe2f76817035e
--- /dev/null
+++ b/train/sample_000024/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000024",
+  "puzzle_id": "o5Csk",
+  "task": "task2",
+  "fen": "8/8/p2k4/1p1p2pp/1P1K4/PP3PPP/8/8 w - - 1 47",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "f4",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000025/board.png b/train/sample_000025/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..a66717afdf6301b7a010600743f76e0807817a93
--- /dev/null
+++ b/train/sample_000025/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:14b86609adf32ab980347272934fb4b44230b3753f540b63db166817661529eb
+size 29377
diff --git a/train/sample_000025/metadata.json b/train/sample_000025/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..caaf84c78aa3d7f8da7b007a8ad23deca38441ab
--- /dev/null
+++ b/train/sample_000025/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000025",
+  "puzzle_id": "sRD7C",
+  "task": "task2",
+  "fen": "r7/8/2p2p1p/2P1k1p1/1P6/P4RP1/5K1P/8 w - - 2 40",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Ke3",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000026/board.png b/train/sample_000026/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..2738f86eee1959039c42929817a1532827ca2723
--- /dev/null
+++ b/train/sample_000026/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d424f983c709e7fdbce463b7a87e6a6b9a9265d90106c1f1685c77f8dbbf9274
+size 36035
diff --git a/train/sample_000026/metadata.json b/train/sample_000026/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..f554833584a8b99827ab4b77eb433d2595177457
--- /dev/null
+++ b/train/sample_000026/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000026",
+  "puzzle_id": "BW0je",
+  "task": "task2",
+  "fen": "4rk1R/5qp1/pp3p2/5P2/2pP4/6QP/P1P5/6K1 b - - 1 37",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Ke7",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000027/board.png b/train/sample_000027/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..318605e3677d32cc88eeee3fe5443defa93eab45
--- /dev/null
+++ b/train/sample_000027/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cc5f4358d316fd69c0ca440643fd648d916fe97f6338cf4965f24a3a57c1ae11
+size 40278
diff --git a/train/sample_000027/metadata.json b/train/sample_000027/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..b01a2059fd1ccab45d27a00da3697e61f7760b0c
--- /dev/null
+++ b/train/sample_000027/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000027",
+  "puzzle_id": "5Tl1t",
+  "task": "task2",
+  "fen": "8/5rpk/p7/1p6/1Pq1p3/3bP1BP/PQ1R1PP1/7K w - - 2 36",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Qe5",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000028/board.png b/train/sample_000028/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..bb16c81a6f1cf9d4346bed574cd8d44a3a29ca2c
--- /dev/null
+++ b/train/sample_000028/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c7128cc9bab9a6d465ecc66c7ff99a59ae1a5376c23ecde5e118c15bd9147216
+size 55676
diff --git a/train/sample_000028/metadata.json b/train/sample_000028/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..8a430d29b0b162bf010eb6e49658b8ca8c062d91
--- /dev/null
+++ b/train/sample_000028/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000028",
+  "puzzle_id": "iXVLq",
+  "task": "task2",
+  "fen": "r3r1k1/1bqn1pbp/p2p1np1/Pp1P4/1PpQPB2/2N2N1P/2B2PP1/R3R1K1 b - - 2 18",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Nxd5",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000029/board.png b/train/sample_000029/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..6e04298e25e94e405c9abd10a96ca650f15f1e64
--- /dev/null
+++ b/train/sample_000029/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac7bf95529543e72412423b7a83218042d1f21d34c5871e5bdae501577a5b978
+size 24225
diff --git a/train/sample_000029/metadata.json b/train/sample_000029/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..7a84a855d247c77e69879e23b38b63fe0fcad775
--- /dev/null
+++ b/train/sample_000029/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000029",
+  "puzzle_id": "muNCa",
+  "task": "task2",
+  "fen": "8/4R3/1P6/3p4/3r4/P3pk2/8/5K2 w - - 3 56",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "b7",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000030/board.png b/train/sample_000030/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..0efb9bf847cd1746e6bba721d0466795260538d0
--- /dev/null
+++ b/train/sample_000030/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4db78ec4c56d37ceeb041c293ee54762b4883a574f6f4c49d2ff8f722364f27b
+size 36167
diff --git a/train/sample_000030/metadata.json b/train/sample_000030/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..3dd25aa265ab2a3aadbff6dd4118bd8181420c0c
--- /dev/null
+++ b/train/sample_000030/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000030",
+  "puzzle_id": "7Xvoi",
+  "task": "task2",
+  "fen": "5r1k/6pp/5p2/p2Qp3/8/2q1P2P/P4PPK/3R4 b - - 11 38",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Rc8",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/train/sample_000031/board.png b/train/sample_000031/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..e1243907f12e7326db75ff127aa0093825b57723
--- /dev/null
+++ b/train/sample_000031/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c3e153346105126892298adf67d02c4827f610354a029986b4420eb17c36ffe
+size 30644
diff --git a/train/sample_000031/metadata.json b/train/sample_000031/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..5dad9a056be4efa7b13af8f6e0a8a17a5081adf1
--- /dev/null
+++ b/train/sample_000031/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000031",
+  "puzzle_id": "rvnIg",
+  "task": "task2",
+  "fen": "8/7p/2pk2p1/p2p1pP1/P2P4/2P1KPP1/8/8 b - - 0 48",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "c5",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/validation/sample_000000/board.png b/validation/sample_000000/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..817c731fa6af8019fa4d0e02194037d4a964eb5a
--- /dev/null
+++ b/validation/sample_000000/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb9bdf1d9cf859c031ad1363c0ad0b94d1516e873cee943ea216eb1e4bb9db90
+size 40221
diff --git a/validation/sample_000000/metadata.json b/validation/sample_000000/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..b5d000ff8627c74db3090bb86ced2f0859f8287d
--- /dev/null
+++ b/validation/sample_000000/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000000",
+  "puzzle_id": "9JiYC",
+  "task": "task2",
+  "fen": "1r2r1k1/5pp1/2b2q1p/4R3/8/2P1Q3/PP3PPP/R1B3K1 w - - 1 21",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Rxe8+",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/validation/sample_000001/board.png b/validation/sample_000001/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..4dbcbd851016f2b756076708a80d3a9b925f5ab7
--- /dev/null
+++ b/validation/sample_000001/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:98d3f944a44bd0edc3a3a00399d56938d19df0ee4972df09498042830ee3caa8
+size 50016
diff --git a/validation/sample_000001/metadata.json b/validation/sample_000001/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..58d0528f727cac6ff5acbd78dc05302590c9b567
--- /dev/null
+++ b/validation/sample_000001/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000001",
+  "puzzle_id": "Lhld2",
+  "task": "task2",
+  "fen": "r4rk1/1bq1bppp/p4n2/1p2p1B1/4P3/1PN2N1P/1P2QPP1/2RR2K1 b - - 3 16",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Rad8",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/validation/sample_000002/board.png b/validation/sample_000002/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..28d3ce0b10400ccd8e40ff05e4a7ccc53fe8524a
--- /dev/null
+++ b/validation/sample_000002/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e8f32d80bea6de8e403dd85fb6745124218f2680ef091f361820a79f440f50d7
+size 43747
diff --git a/validation/sample_000002/metadata.json b/validation/sample_000002/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..d8567a840d7515d0ffaeca2dd34ebb03fa0b44c4
--- /dev/null
+++ b/validation/sample_000002/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000002",
+  "puzzle_id": "RdgvN",
+  "task": "task2",
+  "fen": "r2b1r1k/p5pp/2Qp4/4pP2/2n5/2P3N1/PPq2P1P/3R1RK1 b - - 7 24",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Nxb2",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file
diff --git a/validation/sample_000003/board.png b/validation/sample_000003/board.png
new file mode 100644
index 0000000000000000000000000000000000000000..2663788fa84a8a59737ff0d0621de709f44487a4
--- /dev/null
+++ b/validation/sample_000003/board.png
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:82017c4adbfb3982bb6871923d9fd8025f5017b20888de6c8f27b4e93d34d40f
+size 51849
diff --git a/validation/sample_000003/metadata.json b/validation/sample_000003/metadata.json
new file mode 100644
index 0000000000000000000000000000000000000000..95aec83b7320c2bdfa50b4dcd5c83cace8949669
--- /dev/null
+++ b/validation/sample_000003/metadata.json
@@ -0,0 +1,15 @@
+{
+  "sample_id": "sample_000003",
+  "puzzle_id": "YKK0p",
+  "task": "task2",
+  "fen": "rn3rk1/pb2ppbp/6p1/1ppqP1N1/3P4/4B3/PPP1B1PP/R2Q1RK1 w - - 0 13",
+  "prompt": "You are a specialized model for chessboard understanding.\nYour goal is to predict the last move played on the board based on the highlighted squares.\nInput:\n- Board Image: The visual representation with the last move highlighted.\nOutput Format:\nReturn only the move in Standard Algebraic Notation (SAN).",
+  "target": "Bd3",
+  "image_count": 1,
+  "image_files": [
+    "board.png"
+  ],
+  "image_size": 512,
+  "patch_order": "row_major",
+  "patch_size": 16
+}
\ No newline at end of file